Distribution Category:

LMFBRPhysics: Base
Technology (UC-79d)
A N L 8 3 - 1
A N L
~
8 3
-
1
D E 8 3 0 1 1 0 1 9
ARGONNE NATIONAL LABORATORY
9700 South Cass Avenue
Argonne, Illinois 60439
THE DIF3D NODAL NEUTRONICS OPTION
FOR TWO- AND THREE-DIMENSIONAL
DIFFUSION-THEORY CALCULATIONS
IN HEXAGONAL GEOMETRY
by
R. D. Lawrence
Applied Physics Division
DISCLAIMER
This report was prepared as an?ccount of work sponsored by anagency of the United States
Government. Neither the United States Government nor any agency thereof, nor any of their
er.
M
)loyees, makes any warranty, express or implied, or assumes any legal liability or responsi-
bility for the accuracy, completeness, or usefulness of any information, apparatus, product, or
process disclosed, or represents that its use would not infringe privately owned rights. Refer-
ence herein to any specific commercial product, process, or service bytrade name, trademark,
manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recom-
mendation, or favoring by the United States Government or anyagency thereof. The views
and opinions of authors expressed herein do not necessarily state or reflect those of the
United States Government or any agency thereof.
March 1983
TABLE OF CONTENTS
Paj
1. INTRODUCTION 2
2. DERIVATION OF THE NODAL EQUATIONS IN TWO DIMENSIONS 4
2.1 The Neutron Diffusion Equation 4
2.2 The Nodal Balance Equation 5
2.3 The Transverse Integration Procedure in Hexagonal Geometry . . 8
2.4 Approximation of the One-Dimensional Hex-Plane Equations . . . 13
The One-Dimensional Hex-Plane Polynomial Approximation 14
Calculation of the Expansion Coefficient a|
x
3 17
2.4.1 On the Choice of a Method 13
2.4.2
2.4.3
2.4.4 Calculation of the Expansion Coefficient a|
x
4 21
2.4.5 Overview of the Approximation Procedure 24
2.5 The Flux Moment Equations 25
2.6 The Response Matrix Equation 29
2.7 Boundary Conditions 33
2.8 Summary 34
3. DERIVATION OF THE NODAL EQUATIONS IN THREE DIMENSIONS 35
3.1 The Nodal Balance Equation 35
3.2 The Transverse Integration Procedure in Three Dimensions . . . 37
3.3 Approximation of the One-Dimensional Axial Equation 42
3.3.1 The One-Dimensional Axial Polynomial Approximation . . 42
3.3.2 Calculation of the Expansion Coefficient a|
z
3 43
3.4 The Flux Moment Equations 45
3., 5 The Response Matrix Equation 49
3.6 Calculation of the Leakage Moments 53
3.7 Summary 56
4. NUMERICAL SOLUTION OF THE NODAL EQUATIONS 57
4.1 Overview of the Solution Procedure 57
4.2 Solution of the Response Matrix Equations 60
4.2.1 Two Dimensions 60
4.2.2 Three Dimensions 64
4.3 Coarse-Mesh Rebalance Acceleration of the Outer Iterations 68
4.3.1 Construction of the Coarse-Mesh Equations 69
4*3.2 Solution of the Coarse-Mesh Equations . . 72
ill
TABLE OF CONTENTS (Cont'd)
Page
4.4 Acceleration of the Outer Iterations Using Asymptotic
Extrapolation 74
4.5 On the Computational Efficiency of the Nodal Scheme 76
5. SAMPLE NUMERICAL CALCULATIONS 77
5.1 The SNR Benchmark Problem 77
5.1.1 Two-Dimensional Results 78
5.1.2 Three-Dimensional Results , . . . . 78
5.2 The LCCEWG Benchmark Problem 79
5.2.1 Two-Dimensional Results 82
5.2.2 Three-Dimensional Results 83
5.3 Calculation of Peak Power Densities 85
5.4 Overview of the Numerical Results 86
6. USER INFORMATION 88
6.1 Code-Dependent Input-A.DIF3D 88
6.1.1 Data Management Options and Container Sizes - A.DIF3D
Card Type 02 88
6.1.2 Nodal Option Parameters - A.DIF3D Card Type 10 90
6.1.3 Axial Coarse-Mesh Rebalance Boundaries - A.DIF3D Card
Type 11 90
6.2 Geometry Input - A.NIP3 90
6.2.1 Problem Geometry Specification - A.NIP3 Card Type 03 . . 91
6.2.2 External Boundary Conditions - A.NIP3 Card Type 04 . . . 91
6.2.3 Variable Mesh Structure - A.NIP3 Card Type 09 91
6.2.4 Location of Regions - A.NIP3 Card Type 30 91
6.2.5 Background Region Name - A.NIP3 Card Type 31 92
6.3 Restart Procedure 92
6.4 Limitations of the Nodal Option 93
7. PROGRAMMING INFORMATION 94
7.1 Programming Structure 94
7.2 Data Management Strategy 97
7.3 Additional Data Management Considerations for the CDC 7600
Computer 100
7.4 Additional Programming Details 102
iv
TABLE OF CONTENTS (Cont'd)
Page
7.4.1 Mesh Cell Ordering 102
7.4.2 Partial Current Ordering 103
7.4.3 Storage of the Nodal Coupling Coefficients 103
ACKNOWLEDGEMENTS 105
REFERENCES , . . . 106
APPENDIX A - DERIVATION OF THE TWO-DIMENSIONAL RESPONSE MATRIX EQUATION 109
It
A. 1 Calculation of the Expansion Coefficient a
g x
3 109
A. 1.1 The Relationship between E|
x
( x ) and

g y ( x ) , 109
A. 1.2 Approximation of

g y ( x ) Ill
A. 1.3 Final Form of the Equation for a |
x
3 113
A.2 Calculation of E|
x
( h /2) 114
A. 3 The Response Matrix Equation 114
APPENDIX B - DERIVATION OF THE THREE-DIMENSIONAL RESPONSE MATRIX EQUATION 124
APPENDIX C - APPROXIMATION OF THE LEAKAGE MOMENTS . . . . . 135
C.1 Calculation of the Leakage Moments . . . . . 135
C.2 Model Problem Study of the Errors Associated with the Leakage
Approximations 136
APPENDIX D - CALCULATION OF PEAK POWER DENSITIES 140
APPENDIX E - DESCRIPTION OF THE BCD INPUT FILE A.DIF3D 143
APPENDIX F - DESCRIPTION OF SELECTED CARD TYPES IN THE BCD INPUT FILE
A.NIP3 155
LIST OF FIGURES
Page
2.1 Nodal Coordinate System 5
2.2 Lowest-Order (Quadratic) Hex-Plane Polynomial Approximation . . . 16
2.3 Higher-Order Basis Functions in the Hex-Plane Polynomial
Approximation 16
4.1 Overview of the Nodal Solution Algorithm 58
4.2 The Four-Color Checkerboard Ordering in Hexagonal Geometry . . . 60
4.3 Coarse-Mesh Rebalance Regions in the Hex-Plane . . 69
5.1 Sixty-Degree Sector of the Core Layout for the LCCEWG Benchmark
Problem 80
5.2 Axial Dimensions for the LCCEWG Benchmark Problem 81
6.1 Sample Data Management Edit for the Nodal Option 89
7.1 Subroutine Map for the DIF3D Code Block 95
7.2 GEODST and Nodal Mesh-Cell Orderings (Sixth-Core Symmetry) . . . 102
vi
LIST OF TABLES
Page
5.1 Summary of Results for the Two-Dimensional SNR Benchmark
Problem 78
5.2 Summary of Results for the Three-Dimensional SNR Benchmark
Problem 79
5.3 Summary of BOL Results for the Two-Dimensional LCCEWG Benchmark
Problem 82
5.4 Summary of Depletion Results for the Two-Dimensional LCCEWG
Benchmark Problem 83
5.5 Summary of BOL Results for the Three-Dimensional LCCEWG Benchmark
Problem 84
5.6 Computed Burnup Swings for the Three-Dimensional LCCEWG Benchmark
Problem 84
5.7 Summary of Execution Statistics for the Three-Dimensional LCCEWG
Benchmark Problem 85
5.8 Computed Peak Power Densities for the Three-Dimensional SNR
Benchmark Problem 86
7.1 Scratch Disk Files Used in the Nodal Option 98
7.2 Correspondence Between ECM and Disk Files in the Nodal Option . . 99
7.3 Allocation of Scratch Arrays in thu FCM Array SCRFCM During
Subroutine 0UTR4 (Two-Level Implementation Only) 101
C.1 Two-Group Cross Sections for the Model Problem 136
0.2 Computed Eigenvalues for the Model Problem . . . . . 137
C.3 Error Contributions for the Model Problem 138
vii
The DIF3D Nodal Neutronics Option for Two- and Three-Dimensional
Diffusion Theory Calculations in Hexagonal Geometry
by
R. D. Lawrence
Applied Physics Division
Argonne National Laboratory
Argonne, Illinois 60439
ABSTRACT
A nodal method is developed for the solution of the neutrons-diffusion
equation in two- and three-dimensional hexagonal geometries. The nodal scheme
has been incorporated as an option in the finite-difference diffusion-theory
code DIF3D, and is intended for use in the analysis of current LMFBR designs.
The nodal equations are derived using higher-order polynomial approximations
to the spatial dependence of the flux within the hexagonal-z node. The final
equations, which are cast in the form of inhomogeneous response-matrix
equations for each energy group, involve spatial moments of the node-interior
flux distribution plus surface-averaged partial currents across the faces of
the node. These equations are solved using a conventional fission source
iteration accelerated by coarse-mesh rebalance and asymptotic source
extrapolation.
Numerical calculations for models of heterogeneous-core LMFBR designs
have shown the accuracy of the nodal scheme to be superior to that of the
DIF3D finite difference option with six triangular mesh cells per hexagonal
fuel assembly. The higher-order axial approximation in the nodal scheme permits
the use of an axial mesh which is at least four times coarser than a typical
finite difference mesh. Particular improvement is seen in the average fluxes
in the internal-blanket regions and in the computed values for k-effective,
thus leading to more accurate predictions of internal-blanket burnups,
breeding ratios, and burnup reactivity swings* This enhanced accuracy is
obtained with a potential order-of-magnitude reduction in the computational
cost of a three-dimensional calculation.
This report describes the mathematical development and numerical solution
of the nodal equations, as well as the use of the nodal option and details
concerning its programming structure. This latter information is intended
to supplement the information provided in the separate documentation of the
DIF3D code.
1. INTRODUCTION
The physics and safety analysis of current Liquid Metal Fast Breeder
Reactor (LMFBR) designs requires the cau M l i t y to compute accurate numerical
solutions to the neutron diffusion equation in three-dimensional hexagonal-z
geometry* These neutronics calculations are generally performed within the
U.S. fast reactor program using either standard mesh-centered finite difference
codes
1
""^
o r
flux synthesis methods such as the single-channel flux synthesis
code SYN3D (Ref. 4 ) . Due to the large number of unknowns involved, these
calculations can be very expensive, particularly for fuel management studies
which require repeated solution of the diffusion equation.
At Argonne National Laboratory, depletion calculations using the burnup
cede REBUS-3 (Ref. 5) are performed routinely in support of ongoing fast
reactor design and analysis activities* The depletion calculation in REBUS-3
requires average group fluxes for burnup zones (over which the cross sections
are taken to be spatially constant) defined such that each zone is composed of
at least one hexagonal fuel assembly with an axial zone dimension of approxi-
mately 15 cm* A standard finite difference neutronics calculation requires
six triangular mesh cells per hexagonal fuel assembly and an axial mesh
spacing of approximately 5 cm in order to reduce spatial truncation errors to
an acceptable level. Since only the zone-averaged fluxes are required for the
actual depletion calculation, it is clear that a significant reduction in the
overall computational expense can be achieved by reducing the number of mesh-
points used to approximate the flux in each burnup zone. Thus there exists a
strong motivation to develop a diffusion theory method which will compute
accurate fluxes and eigenvalues when applied on a mesh defined by the dimen-
sions of the hexagonal fuel assemblies and the axial zone boundaries. It is
with this objective in mind that the nodal neutronics module described in this
report was developed. This coarse-mesh neutronics capability is presently
available as an option in the finite difference diffusion-theory code DIF3D
(Ref. 3 ) , which was developed by the Applied Physics Division at Argonne
National Laboratory. The nodal option can also be used to provide the
neutronics solutions required by REBUS-3.
Nodal methods comprise a class OJC coarse-mesh numerical methods which
have demonstrated considerable potential for the analysis of light water
reactors in Cartesian geometry* Many of the earlier nodal schemes
6
involved
empirical coupling parameters which were determined from the results of
detailed fine-mesh calculations or from actual operating data. Nodal schemes
7
developed in the past eight years have, for the most part, eliminated the need
for empirical constants by computing the inter-node coupling relationships
using higher-order approximations to the diffusion equation. Thus, unlike
the earlier ad-hoc methods, these more recent nodal schemes can be viewed as
coarse-mesh approximations to the neutron diffusion equation, and can thus be
expected to converge to the exact solution of the diffusion equation in the
limit as the mesh spacing goes to zero*
The success of these Cartesian-geometry schemes has prompted the more
recent development of analogous techniques
8
"
11
for fast reactor calculations
in hexagonal geometry* The nodal method
1 0
"
1 1
described in this report is
based on a response matrix formulation in which the principal unknowns are the
surface-averaged partial currents across the nodal interfaces. The response
matrix equation is derived using an extension to hexagonal geometry of the
transverse integration procedure widely used in the development of Cartesian-
geometry nodal schemes. Numerical calculations for typical heterogeneous-core
LMFBR designs have shown that the accuracy of the nodal scheme is superior to
that of a standard (6 mesh cells per hexagon, 5 cm axial mesh) finite
difference calculation, and that this improved accuracy is obtained with a
potential order-of-magnitude reduction in the computational cost of a three-
dimensional calculation.
This report consists of two parts. The first part describes the
mathematical development and numerical solution of the nodal equations.
Specifically, Sections 2 and 3 discuss the derivations of the nodal equations
in two and three dimensions, respectively, Section 4 describes the iterative
procedures used to solve these equations, and Section 5 provides some numerical
comparisons between the nodal and finite difference options in DIF3D. The
second part of this report is intended as a user's manual for the nodal option
in DIF3D. Section 6 includes specific information of interest to users of
the code, while Section 7 provides additional information concerning the
programming structure of the nodal option. Since much of the information
provided in the documentation
3
of the finite difference option in D1F3D is
pertinent to the nodal option, Sections 6 and 7 discuss only those additional
features which are unique to the nodal option.
2. DERIVATION OF THE NODAL EQUATIONS IN TWO DIMENSIONS
2.1 The Neutron Diffusion Equation
As stated in the introduction, the objective of this work is to develop
a capability to compute accurate numerical solutions to the neutron diffusion
equation on a mesh defined by the dimensions of the hexagonal fuel assemblies
and the boundaries of the axial burnup regions* Consistent with the present
methodology in the reactor burnup code REBUS-3 (Ref. 5), the cross sections
are assumed to be independent of position within the hexagonal-z mesh cell
(node)* The multigroup neutron diffusion equation
12
for a homogeneous node
V^ can then be written in the form
Z * ( E ) +
, k
4 > < < r), reV
k
, g-l,...,G, (2.1)
where
G
A denotes an eigenvalue, and the remaining notation is standard.
12
Although only the eigenvalue problem is considered here, the applica-
tion of the nodal scheme to fixed-source problems is straightforward*.
Equation (2.1) is solved subject to the boundary conditions that the
flux and surface-normal component of the net current be continuous
across the nodal interfaces, i.e.
*g
(r
s>
"*
D
where r
s
denotes the surface shared by adjacent nodes k and . Boundary
conditions of the general form
a
g g
( r
s
) + 2b
g "*
D
g ^ g
( r
s
)
"
r
s
e S

( 2
-
5 )
*The capability to solve fixed source problems has not been implemented
in the DIF3D nodal option*
are specified on nodal surfaces which form part of the outer boundary S of the
solution domain. Standard boundary conditions (e.g. zero flux, zero incoming
partial current) are obtained via appropriate specification of the constants
a
e
and b
p
in Eq. ( 2.5) .
As will be shown in Section 3, the three-dimensional nodal scheme employs
somewhat different approximations to the spatial dependence of the flux in the
radial (hex-plane) and axial directions. Thus, for the sake of clarity, we
consider only the two-dimensional derivation in this section, and then use
these two-dimensional results in combination with the additional axial
approximations to derive < full three-dimensional nodal scheme in Section 3.
2.2 The Nodal Balance Equation
The starting point in the derivation of the nodal scheme is the nodal
balance equation obtained by integrating the diffusion equation [Eq. (2.1)}
over a homogeneous node V^. Using the orientation shown in Fig. 2.1, with
the origin (in local coordinates) taken as the center of the hexagon, the
k~th node is defined by
(x,y) xe[-h/2,+h/2] , y[-y (x)]
where
y
s
( x ) E ^ (h - ), (2.6)
and h is the lattice pitch. As shown in Fig. 2.1, the u and v directions are
defined as perpendicular to the two sets of opposite faces not perpendicular
to the x-direction.
Fig. 2.1 Nodal Coordinate System
The nodal balance equation is obtained by operating on Eq. (2.1) with
V* J
t
reV
K
where V^ is the volume of the hexagonal node, and then applying Gauss'
theorem to the integrated leakage term:
J A 7-D* V^(r) - $ f
\ Vg
k il k
reV
K x
r eS,
-si
The summation shown here is over the six surfaces of the hexagonal node.
Using the orientation shown in Fig. 2.1, the resulting balance equation
can be written in the form
# [L
k
+ L
k
+ L
k
] + E
r

k

k
- ?, (2.7)
3 gx gu gv
J
g
y
g
x
g*
where the node-averaged values of the flux and the multigroup source term
are defined by
h/2
-k _ j_ f
J 8
-h/2
Y

/
-h/2
and
dx / dy Q
k
(x,y), (2.9)
h/2
VT 2
=j h . (2.10)
k k k
The. terms Lg
X
> ^gu>
an<
* ^gv
a r e
average leakages in the three hex
plane directions, e.g.
f i J*<+h/2) J*" (h/2), (2.11)
gx gx gx

k
where J
gx
(h/2) are surfaceaveraged components of the net current in
the xdirection:
y s < x )
f
dy
"
D
g h ^ y > J x =
h
/ 2
() y
a
(x)
The solution of Eq. (2.7) clearly requires additional relationships
between the surfaceaveraged leakages and the nodal fluxes in the kth node
and its immediate neighbors. It is these additional relationships which
characterize different nodal formulations. As a simple example, consider
the standard meshcentered finite difference equations which are derived
under the assumption that the flux varies linearly from the center of the
hexagon to the midpoints on any of the six surfaces. The resulting
coupling relationships can be written in the form
J
k
(h/2) = Y
k ,
! *
k
Y
, k
*, (2.13)
gx
v
gx+
T
g gx
T
g'
where the coupling coefficients are
, D
k
D*
k & H k 2 E
Y
gx+
= Y
gx"
=
h
D
k
+ D
A '
( 2
g g
Here, I denotes the neighboring node in the positive x-direction such that the
surface at x h/2 is shared by nodes k and . Substitution of Eqs. (2.11)
and (2.13) into Eq. (2.7) yields standard 7-point finite difference equations
in two-dimensional hexagonal geometry. Thus the mesh-centered finite
difference equations can be viewed as a simple nodal approximation in which
coupling relationships of the form given in Eq. (2.13) are derived assuming
a linear flux variation within the node.
The simple form of Eq. (2.13) and the resulting finite-difference-like
form of the equations for the nodal fluxes suggest that Eq. (2.13) may provide
an appropriate basis for more accurate approximations. Su> h improved approxi-
mations can be obtained by using higher-order polynomial approximations to the
spatial variation of the flux within the node. Earlier unpublished work
13
along these lines resulted in a higher-order nodal formulation which utilizes
coupling relationships of the form shown in Eq. (2.13). However, unlike Eq.
(2.14), the expressions for the coupling coefficients involved ratios of
surface-averaged fluxes to node-averaged fluxes as well as the higher-order
coefficients of the polynomial approximation to the flux. This scheme thus
requires non-linear updates of the coupling coefficients during the usual
outer iteration procedure. Another potential drawback is that, unlike the
finite difference matrix, the coefficient matrix obtained in the nodal
scheme is not symmetric. This property, plus the need to update the coupling
coefficients during the outer iteration procedure, raises additional questions
concerning the applicability of the very efficient iterative solution methods
llf
developed for finite difference equations to the nonlinear nodal equations.
Furthermore, numerical studies of the analogous slab geometry scheme
demonstrated that convergence difficulties can arise if the coupling between
the equations for the surface fluxes and higher order coefficients is not
represented properly. Although this latter difficulty was eventually
resolvedj the uncertain iterative convergence behavior of the nonlinear
scheme led to the development of an alternative formulation in which the
inter-node leakages are calculated in terms of interface partial currents*
This linear partial current scheme forms the basis of the DIF3D nodal option.
2.3 The Transverse Integration Procedure in Hexagonal Geometry
The equations for the partial currents required for the evaluation of
the leakages in Eq. (2.7) are derived via an extension to hexagonal geometry
of the transverse integration procedure
7
widely used in the development of
Cartesian-geometry nodal schemes. In Cartesian geometry this technique
involves spatially integrating the n-dimensional diffusion equation over the
n-1 directions transverse to each coordinate direction. The resulting set of
n coupled ordinary differential equations are approximated using techniques
appropriate for the numerical solution of the one-dimensional diffusion
equation. Additional approximations to the transverse leakage terms which
couple the one-dimensional equations are also required.
Direct application of the analogous transverse integration procedure in
hexagonal geometry yields three second-order ordinary differential equations
in the x-, u-, and v-directions. However, a more straightforward procedure
is to derive the P-l forms of these equations using simple neutron balance
arguments. For example, the one-dimensional equation in the x-direction is
obtained by first introducing the partially-integrated quantities
y
s
(x)
* < x)
8
= I dy / ( x , y ) (2.15)
J g
y
s
(x)
J
k
gx
(x) s J dy - D J | J
J
- y
s
(x)
/
y
s
(x)
s
dy Q*(x,y), (2.17)
and then performing a simple neutron balance on the line defined by
6V
k
: (x,y) xe [x, x+dx], ye[-y(x),+y (x)],
s s
The resulting balance equation can be written in the form
where Jg(x,y
s
(x)) are surface-normal components of the net current
across the u- and v-directed surfaces:
"g -
+
- l ; (.) I , . ,
W
< 2-19a)
n.-Z .(x> y) I
T
. ^
(
. (2.19b)
s
10
Here, n
+
denotes unit vectors normal to the u- and v-directed surfaces
n. =
n
v+
n
u+
-h/2 < x < 0
0 < x < h/2
(2.20a)
n =
n
u-
n
v-
-h/2 < x < 0
0 < x < h/2
(2.20b)
where, for example, 1^+ is the unit vector normal to the surface in
the positive u-direction shown in Fig. 2.1. As shown in Section 2.5,
integration of Eq. (2.18) over xe[-h/2, +h/2] yields the nodal balance
equation, Eq. (2.7), as it should.
It is also convenient to introduce for later use the y-averaged
quantities

w
7
e
(x>
2y
g
(x)
(2.21)
and
.(X) 5
f
y
e
(x)
dy
'
D
g n
(2.22)
11
Sxnce Eq. (2.18) is written in P-l form, we require an additional
equation (analogous to Fick's.Law) relating the partially-integrated
flux < |> g
X
(x) and net current J
g x
( x ) . This relationship is obtained
by applying Leibniz
1
rule for differentiating an integral with variable
limits to Eq. ( 2.15) :
y
s
(x)
~k d ,k , v .. _k d I , ,k, v
-D $ (x) = -D -r- I dy 4> (x,y)
g dx
Y
g x
v
g dx J / f
g
v ,//
_y
s
(x)
dy
a ^
- D^ y'(x) [^(x,y(x)) + ^(x,-y (x))]. (2.24)
g s g s g s
Since the first term on the right hand of this last equation is simply
Jg
X
(x), rearrangement yields
J
g V.
(x) =
~
D
g L
< |,
gx
(x) +
g
y
s
( x )
t *g <
x
y
s
^
+
< Pg(x,-y
s
(x))]. (2.25)
-k k
Similar
,
y, the following relationship between <j>g
X
(x) and J
gx
(x) is
obtaiied '
J (x) = - D
K
(J)
k
(x) t D
K
0
S
. . E
K
(x), (2.26)
g x
v
g dx
y
g x
v
g 2y (x) g x
v
' '
where
E
k
(x) H ^(x, y (x)) + <(x,-y (x)) - i f -
v
(x). (2.27)
g X g o 6
s,
6
X
lr
The terms 4> g(x,y
s
(x)) are fluxes evaluated on the u - and v-directed
surfaces.
12
Equation (2.25) is similar to Fick's Law, but includes an additional
term involving fluxes on the u- and v-directed surfaces* This additional
term, which does not occur in Cartesian geometry, results from the variable
integration limits inherent in hexagonal geometry. Differentiating Eq.
(2.6) yields
y!(x) - - ^s g n ( x ) (2.28)
s yj
(2.29)
where 6(x) is the Dirac delta function. Since y
g
(x) is discontinuous
at x"0, the second term on the right hand side of Eq. (2.25) exhibits
this same behavior. However, consistent with Eq. (2.4), the partially-
integrated x-component of the net current must be continuous at x
_
0.
Therefore, the partially-integrated flux must exhibit the following
first-derivative discontinuity at x0:
lim [-D
k
3- <|>
k
( x) ]
X
"

& [*
k
(x,y ( x) ) + <J>
k
(x,-y ( x) ) ]
A
. (2.30a)
0
g dx *gx ' *x-e y j -
y
g '
J
s g r> x0
The y-averaged flux exhibits a similar discontinuity:
D
k
H > T - (x)]
e
- T-* E (0). (2.30b)
e-K)
8 g x x
*~
e g x
This behavior must be represented by any polynomial used to approximate
the one-dimensional fluxes in hexagonal geometry.
The u- and v-direction counterparts to Eqe. (2.18) and (2.25) are
derived in an analogous manner.
The approximation techniques developed in the following section are
applied to the P-l form of the one-dimensional equations given by Eqs.
(2.18) and (2.25). However* ir. order to facilitate comparison with the
usual second-order differential form of the one-dimensional Cartesian-
geometry equations, it is convenient to cast Eqs. (2.18) and (2.25) in
this same form. This result is obtained by substituting Eq. (2.25) into
Eq. (2.18), and then using Eqs. (2.28) and (2.29):
13
where
D
k
f sgn(x)
ldx

6 (X ) [<Ox,y (X )) + < T(x,-y(x))]. (2.32)
g s g s
Although Eq. (2.31) is of the same general form as the Cartesian-geometry
equations (i.e. a one-dimensional diffusion equation with a modified source
term accounting for leakage in the transverse direction), the expression
for the transverse-leakage term S
x
(x) is considerably different. In
particular, Eq. (2.32) includes two additional terms involving the Dirac
delta function 6(x) as well the Eluxes and their derivatives evaluated on
the surfaces of the nodes. The impact of these additional terms on the
choice of an approximation scheme is discussed in the following section.
2.4 Approximation of the One-Dimensional Hfcx-Plane Equations
2.4.1 On the Choice of a Method
A number of methods 15-20 h a
Ve
been developed for the approximate
solution of Eq. (2.31) in Cartesian geometry. One possible classification
of these methods is on the basis of whether information obtained from an
analytic solution of the diffusion equation within the node is incorporated
into the numerical scheme. In the first class, we include schemes in
which the one-dimensional partially-integrated fluxes are approximated
by a polynomial without the use of analytic information. Examples of
these polynomial methods are the nodal expansion method
1 5
(NEM), the
polynomial scheme developed by S i ms ,
1 6
and the NODLEG method due to
Maeder.
1 7
Examples of the second class, the analytic methods, are the
QUANDRY method,
18
the AN2D method,
19
and the nodal Green's function
method
20
(NGFM).
The distinction between the polynomial and analytic approaches is
particularly relevant to the solution of the transverse-integrated equa-
tions in hexagonal geometry. The QUANDRY ancLAN2D methods solve Eq. (2.31)
(in Cartesian geometry) by first projecting S
g X
(x) onto a low-order poly-
nomial, and then solving the resulting equation analytically. (The AN2D
method also projects Q
g x
(x) onto a low-order polynomial). Hence the
treatment of the delta function contribution in S
g x
(x) would appear to
require rather extensive reformulations of these schemes. Since the NGFM
14
solves Eq. (2*31) via the use of a Green's function for the one-dimensional
diffusion-removal operator on the left hand side of Eq. (2.31), the appli-
cation of this scheme to the solution of Eq. (2.31) is relatively straight-
forward. Numerical r e s u l t s
1 8 - 20
have shown that the analytic methods are
capable of very high accuracy when applied on a mesh corresponding to the
dimensions of the (homogenized) fuel assemblies in a light water reactor
(LWR). However, it should be noted that, measured in diffusion lengths, a
typical LMFBR fuel element is smaller than a LWR fuel assembly; hence the
high accuracy of the analytic methods may be unnecessary for the solution
of Eq. (2.31) for LMFBR applications.
The expansion coefficients in the polynomial methods are calculated
by requiring that the one-dimensional polynomial satisfy Eq. (2.31) in a
weighted-integral sense. Thus the treatment of the delta function is
straightforward, provided that the resulting first-derivative discontinuity
in 4> gx(x) is represented properly by the approximating polynomial.
Since the high accuracy of the analytical methods is probably
unnecessary for our application, and the delta function contribution is
more easily accommodated by the polynomial methods, an approximation
scheme based on a polynomial approach has been developed for the solution
of the one-dimensional equations in hexagonal geometry. Although this
scheme could be applied directly to Eq. (2.31), we choose instead to
approximate the equivalent P-l form [Eqs. (2.18) and (2.25)] since the
resulting derivation is somewhat more straightforward.
2.4.2 The One-Dimensional Hex-Plane Polynomial Approximation
The polynomial approximation to the one-dimensional flux <|> (x)
is given by
N
*gx
(
'
c)
" *g x
( x ) E 2 y
s
( x ) [
*g
+
]
a
gxn
f
n
( x ) ]
'
2
*
N
<
4
'
(2
'
33)
n-1
whei ^
a
gxl
E
*g x
( +h / 2)
" *g x
(
"
h / 2) ( 2 , 3 4 a )
a
gx2
E
*g x
( +h / 2) +
*g x
(
~
h / 2)
"
2
*g
( 2
'
3 4 b )
fj(x) s ^ H K (2.35a)
15
f
2
(x) = y | C
2
- (2.35b)
f
3
(x) = i |c
2
- | |C| +~ (2.35c)
f
4
(x) = C(|e| - j). (2.35d)
The polynomial approximationgiven in Eq. (2.33) is constructed such
that the node-averaged flux <j>g and the surface-averaged fluxes
(}>g
X
(h/2) are preserved, i.e.
i.
v
k
h/ 2
dx <| > ( x) = <j > ( 2. 37a)
h/2
x=h/2
S
*g
Consistent with Eqs. (2.37), the basis functions defined in Eqs. (2o35)
satisfy r.he constraints
h/ 2
dx 2y (x) f (x) = 0, n=1 4; (2.38a)
s n
-h/2
f (h/2) = 0, n=3.4. (2.38b)
n
Note that Eq. (2.33), upon division by 2y
g
(x), reduces for N*=2 to a
quadratic polynomial uniquely determined by the three constraints given
in Eq. (2.37). This lowest-order approximation, which is equivalent to
a quadratic approximation to the y-averaged one-dimensional flux defined
in Eq. (2.21), is illustrated in Fig. 2.2.
16
Numerical studies have shown that while the N=2 approximation is
more accurate than the 1 point-per-hex finite difference approximation,
the quadratic results are not as accurate as the standard 6 triangles-
per~hex finite difference method* One reason for this relatively poor
accuracy is that the N=2 approximation does not represent the first-
derivative discontinuity [Eqs. (2.30)] at x=0. It can be shown that
H./2)
-h/2
i
0
+h/2
Fig. 2.2 Lowest-Order (Quadratic) Hex-Plane Polynomial Approximation
fix)
tin)
-h/2
+h/2
Fig. 2.3 Higher'-Otder Basis Functions in the
Hex-Plane Polynomial Approximation
17
ignoring this behavior introduces an 0(h
2
) error in the calculation;
thus we should not expect the N-2 approximation to be significantly more
accurate than the 1 point-per-hex finite difference scheme which also
converges as 0(h"
:
). The higher-order basis functions f~(x) and f*(x)
were thus added in order to provide an approximation to the first-
derivative discontinuity, and to improve the overall accuracy of the
nodal approximation. These basis functions are plotted In Fig. 2.3.
From Fig. 2.3 it can be seen that f,(x) has a first-derivative dis-
continuity at x= 0, and is thus intended to provide an approximation to
the corresponding behavior in < |> g
X
(x). The basis function f/(x) provides
a quadratic approximation within each of the half-intervals xe[-h/2,0]
and xe[0,h/2], and thus offers improvement in the overall accuracy,
provided that the expansion coefficient aH
x
4 is computed in an appro-
priate manner.
2.4.3 Calculation of the Expansion Coefficients a
E
x3
The coefficient a
g x
3 in Eq. (2.33) is calculated by requiring
that 4>
gx
(x) satisfy Eq. (2.30a), i.e.
lim [-D $ (x)] = lim [-D -j d> (x)j
8 d x g x x =
"
e
e-K)
8 d x g x x =
~
e
2D
k
(2.39)
This is equivalent to requiring that the partially-integrated net current
j|
x
(x) be continuous at x= 0. Differentiating Eq. (2.33) yields
2y(
x
)
Urn t-D
k
%- ^ ( x) ]
X = e
= -D
k
[2y (x) a
k
f(x) +
9
8
f
. *
k
( x) ]
X
"
e
* g dx
y
gx x=-e g
/
s
v
gx3 3
V 7 2
y
g
<
x
) gx x=-e
2D
k
ta
gx3
+ 2
where *g
X
(x) is defined in Eq. (2.21). Substituting Eq. (2.40) into
Eq. (2.39) and then solving for a|
x
3 yields
V 3
5 E
g
18
k k
where E
g x
( x ) is defined in Eq. (2.27). Thus the calculation of a
g X
3
reduces to the development of an approximation to E^
x
( x ) .
k
At first glance, the approximation of E
x
( x) does not appear straight-
forward due to its dependence upon the surface fluxes 4> g(x,y
s
(x)). For
example, EJJ
X
(O) involves the point fluxes at the intersections of the u-
and v-directed surfaces in Fig. 2.1. However, we note that Eq. (2.27),
upon division by 2y
s
(x), is of the same form as the familiar finite
difference approximation to the second derivative. This in turn suggests
that E|
x
(x) can be related to the y-directed leakage defined by
*
k
.
gy
W
s
( x )
^ f
Indeed, such a relationship does exist, and is given by
E
L <
X
> IT [2y(x)l
2
* (x) + 0(h*). (2.43)
gx
6 D
K s gy
g
This result, which is derived in Section A.1.1 of Appendix A, is
particularly welcome since the approximation of the transverse leakage
term f|
y
(x) is relatively straightforward.
A transverse leakage term analogous to Eq. (2.42) arises in the
derivation of Cartesian-geometry nodal schemes. The simplest approxi-
mation to this term is to replace it by its average value over the node.
In Cartesian geometry the average value of the y-directed leakage is given
by the difference of the surface-averaged values of the net current on the
two y-directed surfaces of the node. A more accurate (and thus more
popular) approximation is obtained by replacing ijSy(x) by a quadratic
polynomial
15
written in terms of average leakages in the k-th node and
its two immediate neighbors in the x-direction.
In hexagonal geometry, we replace *gy(x) by the "two-step" approximation
(
-h/2 < x < 0
gy-
(
gy-
(2.44)
where
19
gy-
c r
y
s
(x)
-y
s
(x)
3
2
.k.
(2.45a)
gy+
=
/
/,
h/2
(x)
k3
k k k
It is clear that

g y - and
g
y + are simply average values of
g
y(x)
over the respective half-noae intervals.
The half-node averages defined in Eqs. (2.45) are calculated
in the following manner. Subtracting Eqs. (2.19) yields
- D
k
[n -V
g L + "
<Cg(x,y)
n - V
yy
.00-
y=y
g
(x)
(2.45b)
(2.46)
Explicit evaluation of the n*V terms yields
J
k
(x,y
s
(x)) - J
k
(x,-y
s
(x)) - - D
k
U s g n (x)
7 ^7 *(
x
. y )
Jy~-y (x) f
(2.47)
The total and partial derivatives are related by
* y ' OO ! (2.48)
20
Substituting Eq. (2.48) into Eq. (2.47), and then using Eq. (2.28) yields
/(x,y
s
(x)) - /(x,-y
s
(x)) - -D* U sgn(x) [^ ^(x,y
g
(x))
dx
With reference to Fig. 2.1, we note
1
that
2
f
"
J
-h/2
/
dx [Jg(x,y
s
(x)) - Jg(x,-y
s
(x))] s Jg
y
(h/2) - jJ
u
(-h/2) (2.50a)
/
h/2
dx fJ^(x,y
s
(x)) - Jg(x,-y
s
(x))] = Jg
U
(+h/2) - J^(-h/2), (2.50b)
0
k ~k
where J
g U
(

h /2) and J
g V
(
n
/ 2) , the surface-averaged net currents
across the u- and v-directed surfaces, are defined in analogy with
Eq. (2.12). Performing the y-integration in Eqs. (2.45) yields
f dx U ^ J- Ax , y ) l
ymys
* - I V
k

k
(2.51a)
-h/2
J
s
h/2 r
dx -D^ I- <(x,y) I " - T f ** . (2.51b)
An equation for

g y + is obtained by substituting Eq. (2.49) into Eq*
(2.50b), and then using Eq. (2.51b) to eliminate the final term in Eq.
(2.49) in favor of

gy+ An analogous procedure is used to obtain an
equation for

gy- These results, plus details of the derivations,
are given in Section A.1.2 of Appendix A.
21
With these results in hand, the final equation for the expansion
coefficient aj>
x
3
c a n n o w D e
derived. This final stage of the derivation
proceeds as follows:
a
g x3
= E
g x
( 0 )
l
f r o m Eq
E
g
X
( 0 + ) ]
[2y (0 )]
2
[** + i .
k
} {using Eq. (2.43)}
- 2 [Z* + 4T J. (2.52)
9 D
k gy- gy+
g
Note that E
x
(0 ) is obtained by averaging the values on either side
of x=0 since EJ
x
(x) is not continuous at x=0. [This is due to the two-
step approximation of
f
gy(x) given by Eq. (2.44)]. The 0(h'
f
) terra in
Eq. (2.43) has been neglected, but this error is clearly small compared
r.o that introduced by the two-step leakage approximation. As shown in
Section A.1.3 of Appendix A, Eq. (2.52) leads to the following final
form of the equation for a^ -:
If l *^< */ + g
x
< -"/2) - 2 $ . (2.53)
7
g
As will be shown in Section 2.6, the fluxes and leakages on the
right hand side of Eq. (2.53) are eliminated in favor of interface
partial currents and spstial moments of the intra-node group source
distribution. Thus the coefficient a^ o does not appear in the final
form of the nodal equations.
2.4.4 Calculation of the Expansion Coefficient
The expansion coefficient ax4
i s
calculated using a weighted
residual (WR) approximation to the one-dimensional balance equation,
Eq. (2.18). The WR equation is obtained by weighting Eq. (2.18)
22
with weight function w (x), and then requiring the result to be zero when
integrated over the interval x e[-h/2,+h/2], i.e.
+ ,= U
k
(x,y (* - ^(x.-y (xl> 5 o, (2.54)
V3 8
s
8
s
where the inner product is defined by
1
v" /_
, . dx w(x) $
8* ,,
k
1 S
x
-h/2
Using Eqs. (2.50), it can be shown that unit weighting, i.e.
w(x) = w
Q
(x) 2 1, (2.55)
reduces Eq. (2.53) to the nodal balance equation, Eq. (2.7). Since the
nodal balance equation insures a neutron balance over the hexagonal node,
a logical choice for an additional weight function wj(x) is such that a
neutron balance is preserved over each of the three pairs of half-nodes
in the three hex-plane directions x, u, and v. This is accomplished in
the x-direction by specifying
w(x) = Wj(x) = sgn(x) (2.56)
in Eq. (2.54). That this procedure is equivalent to preserving a neutron
balance over each half-node can be verified by first writing Eq. (2.54)
explicitly for w(x) = w^(x) and w(x) = w , (x),
1 r r
h / 2
< w
Q
( x ) , ... > = [ I dx ... +1 dx ...] = 0
.h/2
'-h/2 " 0
1 c r
h / 2
,... > = - r [- I dx ... +1 dx ...] = 0,
23
and then separately adding and subtracting these results to obtain the
half-node balance equations
f
^r I dx ... = 0 (2.57a)
V J
-h/2
v
k
J
h/2
dx ... =0 . (2.57b)
0
An analogous weighted residual procedure, applied to the u- and v-
direction analogs of Eq. (2.54), insures a neutron balance over each
of the half-nodes in the u- and v-directions.
An equation for aH
x
4 is derived by requiring the one-dimensional
polynomial approximation [Eq. (2.33)] to <|>g
X
(x) satisfy Eq. (2.54) with
w(x) given by Eq. (2.56). Introducing the x-direction spatial moment,
v
k
J
h/2
dx sgn(x) 4>
k
x
(x) (2.58a)
-h/2
h/2
(2.58b)
V
D
'-h/2
i r
h / 2
r
s
k
- j ^ I dx s g n (x) J dy < J>
g
(x,y),
and then substituting Eq. (2.33) into Eq. (2.58a) and performing the
necessary integrations yields
y
gxl 9 gxl 24 gx4*
24
or, using Eq. (2.33a),
a
gx4 " ~
24
W
+
I *
(+h /2
> *
Using this result, ajg
X
4 can be eliminated in favor of the flux moment
$exl which, along with the nodal flux <j>g and the u - and v-direction
moments < j> g
U
l and ^evl> comprise the flux moments which appear in the
final form of the nodal equations* Equations for the flux moments are
derived in Section 2.5.
2.4.5 Overview of the Approximation Procedure
It is appropriate at this point to provide a brief overview of the
approximation procedure developed in this section* The partially-integrated
flux is approximated by the polynomial given in Eq. ( 2.33) . In its simplest
form this approximation reduces to a quadratic polynomial derived such that
the node-averaged flux and the two surface-averaged fluxes in the x-direction
are preserved. [This quadratic approximation is analogous to the lowest-order
approximation used in the Cartesian-geometry Nodal Expansion M e t h o d
1 5
(NEM)] .
Higher-order approximations are obtained by first adding a basis function
f3(x) which has a first derivative discontinuity at x= 0, and then adding an
additional basis function f4(x) which provides a quadratic approximation over
the half-intervals -h/2 < x < 0 and 0 < x < h /2. The coefficient of f3( x) is
determined by requiring the y-integrated net current to be continuous at x*0,
while the coefficient of f4(x) is calculated by applying a weighted residual
procedure to the one-dimensional balance equation. This latter procedure is
equivalent to enforcing a neutron balance over each of the half-nodes. The
calculation of the coefficient a g
x
3 also requires an approximation
[Eq. (2.44)] to the transverse-leakage term ^y ( x ) introduced in Eq. ( 2.42) .
The approximations introduced in this section can also be viewed in a
more general context. The surface-averaged fluxes and leakages are eventually
eliminated in favor of surface-averaged partial currents in the derivation of
the response matrix equation given in Section 2.6. The partial currents are
required during the global solution procedure to be continuous across the
nodal interfaces. This is equivalent to requiring that the surface integrals
of the flux and the surface-normal component of the net current be continuous
across the interface. Furthermore, the nodal balance equation (which is
obtained by integrating the diffusion equation over a node) insures a neutron
balance over the hexagonal node. Thus the following constraints are satisfied
over the hexagon: (i) node-integrated neutron balance, (ii) continuity of the
surface-integrated flux, and (ill) continuity of the surface-integrated
surface-normal component of the net current. These constraints, which are
satisfied by both the quadratic and the higher-order approximations, are
equivalent to requiring that Eqs. ( 2.1 ) , ( 2.3 ) , and (2.4) be satisfied in an
integral sense. As discussed in the preceeding paragraph, the higher-order
expansion coefficients are calculated such that the partially-integrated net
current is continuous across the interface shared by the two half-nodes and
25
such that a neutron balance is preserved over each half-node* Since the
partially-integrated flux is continuous across the half-node interface
(Eq. (2.33) is continuous at x0), the above constraints are also satisfied
over each pair of half-nodes in the three hex-plane directions. In other
words, if the hexagon is divided into six equilateral triangles (by drawing
connecting lines between the three pairs of opposite vertices), the three
constraints are satisfied over the six regions defined by the union of two
adjacent triangles sharing a common surface. (It should be noted, however,
that these constraints are not necessarily satisfied over each of the six
triangles). Thus, in summary, the approximations introduced in this section
result in the diffusion equation [Eq. (2.1)] and the usual continuity con-
ditions [Eqs. (2.3) and (2.4)] being satisfied in an integral sense over each
of the three pairs of half-hexagons as well as over the hexagonal node itself.
2.5 The Flux Moments Equations
The flux moments equations are derived from Eq. (2.54), which we
write in the form
<w (x), 3 - J
k
(x)> + Z
r , k
<j>
k
= Q
k
n dx gx g gxn gxn
- < w (x), p. [J
k
(x,y (x)) - J
k
(x,-y (x))]>, n=0,1, (2.60)
n y^ g s g s
where
.k _ rk
Q
gx0 = V
and w
Q
(x), w ^ x ) , and <j)|
x
i are defined by Eqs. (2.55), (2.56), and (2.58),
respectively. Integrating the first term in Eq. (2.60) by parts yields
< w
n
( x )
> k
J
g
[w
n
(0
+
) - w
n
(0 ")]. (2.61)
Substitution of Eqs. (2.55) and (2.56) into this result yields
26
and
< w.(x), 4 - J
k
(*)> -Ir-T ^ " lu J
k
(0), (2.63)
l
v
* dx gx 3 gx 3h gx
v
"
v
'
-k
respectively, where L
g x
is the x-directed leakage defined in
Eq. (2.11), i.e.
(2.11)
and
Using Eqs. (2.50), the last term in Eq. (2.60) can be written for
n= 0,1 as
t
J ( + h / 2 )
"
3
g u
(
"
h / 2) + 3
g u
( + h / 2 )
"
and
(x,y (x)> J( x , y
8
s
8
s
[
~
3
g v
( + h / 2 ) + 3
g
U
(
-
h / 2 ) + J
g u
( +h / 2
> - Jg
27
Substitution of Eqs. (2.62) and (2.65) into Eq. (2.60) (written for
n=0) yields
r
'
k
$
k
= Q
k
- [L
k
+ L
k
+ L
k
]. (2.67)
g
T
g ^g 3h gx gu gv
Thus, as mentioned previously, unit weighting of the one-dimensional
balance equation yields the nodal balance equation given in Eq. (2.7).
Substitution of Eqs. (2.63) and (2.66) into Eq. (2.60) (written for
n=1) yields
k
, ~ 4- [T
k
+ f
k
- f
k
] + |- J
k
(0). (2.68)
gxl 3h gx gu gv' 3h gx
v
Evaluation of the final term in Eq. (2.68) requires Eq. (2.26), which
we repeat here
J (x) = -D
K
5_ f (
x
) +
D
K
-
s
E
K
(x). (2.26)
gx
v
g dx
v
gx
v
g 2y (x) gx
v
k
Using Eq. (2.33) to approximate <j> (x) yields
- D
k
T
k
/ M _ 8 r k , 7 2 k
r
. k , 2 0
J
gx
( 0 ) =
h
[a
gxl
+
"13
a
gx2
5 + a
gx3
(
T3
- D
k
g
r
k l k l k
h
[ a
g x l - 2
a
-
a
v 1 V
[ a
gxl " I
a
gx4l {using Eq. (2.41)}
- 12 ^
x l
] . {using Eq. (2.59a)| (2.69)
28
Substitution of Eq. (2.69) into Eq. (2.68) yields
[2
r,k
+
32 V k k 2
[f
k jk _-k
] +
4 0 V k
1
g h h
Y
g xl
s
g xl 3h
l
gx gu gv
J
9h h gxl
(2.70)
The u - and v-direction analogs of Eq. (2.70) are readily derived
via the transformations:
u-direction:
v-direction:
, u+v, vx
, u +- x, vu.
(2.71)
The resulting equations can be combined with Eq. (2.70) to yield
k 1
P
gxl
.k
'gul
a
Q
gxl
k
.<tl.
2
3a
k
1
1
1
- 1
1
1
1
g
T
k
gv-i
r k
3
gul
k
a ,
. gvl
(2.72)
where
k r k e
a , = bZ * + 32 T ^ .
gl g h
(2.73)
Rewriting Eq. (2.67) yields
29
Q
k
^-p [L
k
+ L
k
+ L
k
]. (2.74)
8
3hE
r
'
k 8 X g U g V
g
Equations (2.72) and (2.74) are used to evaluate the flux moments
^ ^xl 't'l
an(
* ^ which along with the r fe a v e r g e d g g 't'gul*
an(
* ^gvl* which, along with the surface-averaged
partial currents introduced in the following section, form the
principal unknowns of the nodal scheme.
2.6 The Response Matrix Equation
In this section the approximations introduced in Section 2.4 are
used to derive the local response matrix equation which forms the
cornerstone of the DIF3D nodal scheme. This equation relates the six
outgoing surface-averaged partial currents for a single energy group
to the six incoming partial currents for the same node and the spatial
moments of the intra-node group source distribution. The outgoing and
incoming surface-averaged partial currents across the x-directed faces
shown in Fig. 2.1 are defined by
y
s
( x )
d y
*
-y
g
(x)
y
s
( x )
T
f
dy
ti *> > ? )
+
- 1
D
g h *g
( x
'
y ) }
J
x = +h / 2
( 2
-
7 5 b )
V
X )
The partial ci-rrents across the u- and v-directed faces are defined
in an analogous manner. The approximations developed in Section 2.4
involved both surface-averaged fluxes and net currents. These
quantities are eliminated in favor of the partial currents using the
simple relationships
k
(h/2) = 2[J
U
>
k
(h /2) + J^'
k
(h/2)] (2.76)
gx gx gx
f (+h/2) J
out
'
k
(+h/2) - J
ln,k
(+h/2) (2.77a)
gx gx gx
29
(2.74)
Equations (2.72) and (2.74) are used to evaluate the flux moments
^g
<
l'gxl> '''gul*
an(
* ^gvl* which, along with the surface-averaged
partial currents introduced in the following section, form the
principal unknowns of the nodal scheme.
2.6 The Response Matrix Equation
In this section the approximations introduced in Section 2.4 are
used to derive the local response matrix equation which forms the
cornerstone of the DIF3D nodal scheme. This equation relates the six
outgoing surface-averaged partial currents for a single energy group
to the six incoming partial currents for the same node and the spatial
moments of the intra-node group source distribution. The outgoing and
incoming surface-averaged partial currents across the x-directed faces
shown in Fig. 2.1 are defined by
y
s
( x )
J- .
k
(
h /2)
S
[ ^ f dy (I ,
k
(x,y) , I D
k
-y
e
(x)
( h
/ ) l ^ j f dy {i
The partial currents across the u- and v-directed faces are defined
in an analogous manner. The approximations developed in Section 2.4
involved both surface-averaged fluxes and net currents. These
quantities are eliminated in favor of the partial currents using the
simple relationships
t (h/2) 2[J"
t

k
(h/2) + J^
k
(h/2)] (2.76)
gx gx gx
^ ( h / 2) - J"
t

k
(+h/2) - J^
k
(-Hi/2) (2.77a)
gx gx gx
30
J
k
(-h/2) - J^'Vh /2) - Jf'
k
(-h/2). (2.77b)
g x g x gJi
These relationships are derived by separately adding, and subtracting
Eqs. (2. 7 5), and then recalling the definitions of <f>g
X
(x) and J
g x
(x)
given in Eqs. (2.21) and (2.22).
An equation for J
o u t

k
(+h /2) is derived using Eqs. (2.77a) and
(2.26):
g x
J
o u t , k
(+h /2) - J
k
(+h/2) + J
i n , k
(+h /2)
gx gx gx
v'(x)
f-D
K
%- <J.
K
(x) + D
K
-rr-T -s
E
(x)] ui o + J ' (+h/2)
g dx
y
g x g 2y (x) gx
v
'x^h/t gx
D
k
The first term on the right hand side of Eq. (2.78) Is approximated
using the polynomial approximation given in Eq. (2.33):
N
[-D
k
L. l
k
(
x
)] s - D
k
T - [*
k
+ y a
k
f (x)l .
lo
g dx
v
g x x=h/2 g dx
i y
g L*t gxn n
v
x=h/2
n=1
- D
k
g
r
k 36 k 7 k j . l k ,
/o 7 Q
.
rf- la , + T T a + -rr a + -s- a . J. (2.79)
h gxl 13 gx2 26 gx3 2 gx4
Substitution of Eq. (2.79) into Eq. (2.78) yields
e . k , ^p_ k , 7__ k , _ k
gx V"
1
""/'./ ^ L g
X
j i3 g
X
2 26 gx3 2 gx4
(h/) (2.80)
gx
31
Since the remaining steps in the derivation of the response matrix
equation involve a considerable amount of algebraic manipulation, these
steps are simply outlined here with additional details provided in
Section A.3 of Appendix A. The remaining steps are as follows:
(1) The expansion coefficients a.
k k j k
a
gxl a
g
x2 a
g
x3 and a
g x
4 are
eliminated from Eq. (2.80) using Eqs. (2.34a), (2.34b), (2.53), and
(2.59b), respectively. The term E^
x
(h/2) is eliminated using Eqs.
(2.43) and (2.44).
5
(2) The flux moments
eliminated in favor of
(2.74) and (2.70), respectively.
k k
<j>g and < |> g
X
i introduced via step (1) are
tne source moments Q
o
and Q H
V
I using Eqs.
Qg and Q|
x
i using Eqsi
(3) All surfaced-averaged fluxes and net currents introduced via steps (1)
and (2) are eliminated in favor of surface-averaged partial currents using
Eqs. (2.76) and (2.77). The result, which involves outgoing and incoming
^k **$
partial currents on the six surfaces; plus the source moments Qg and
can be written in the form
Ea
l
a
2
a
3
a
4
a
3
a
2
]
4
out,k
b
2
]
-I
T
i n , k
c
2
] J
g
(2.81)
where J
u
* and J * are column vectors containing the six out-

g "8
going and six incoming partial currents, respectively, for the
k"th node, e.g.
J
U t
'
k
g
3 col
gx
t
J"
t
'
k
(+h/2), J
gv gx
3
o u t , k
(
_
h / 2) >
j
(2.82)
The constants a , b , and c are given in Appendix A.
(4) Five additional equations similar to Eq. (2.81) can be obtained
by applying steps (1) through (3) to Eq. (2.80) written for the remaining
5 surfaces of the hexagonal node* These equations can also be obtained
by applying successive 60 rotational transformations to Eq. (2.81). Note
32
that the u- and v-direction analogs of Eq. (2.81) involve Q|
u
i and Qg
V
l
respectively. These additional aquations can be combined with Eq. (2.81)
to yield
[A
k
] j
o u t
'
k
- [
B
k
] 2
k
+ [C
k
] J
i n , k
, (2.83)
o o o o o o
where
Q
k
H col [Q
k
, Q
k
,, Q
k
,, Q
k
, ] . (2.84)
g g gxl'
x
gul*
x
g vl
J
The structure of the matrices introduced in Eq. (2.83) is discussed
in Appendix A.
(5) The final form of the response matrix equation is obtained by
inverting [A|] in Eq. (2.83) to yield |
p
kj k
+ [R
kj jin.k (2.85)
g g g "g
k k
where [Pg] is a 6 by 4 source matrix and [R
g
] is a 6 by 6 response
matrix.
Equation (2.85) is an inhomogeneous local response matrix equation
written for the k-th node and g-th energy group. The energy groups are
coupled via the source term Q
k
which is calculated [as in Eq. (2.2)]
in terms of the flux moments:
G
k
l ^ " ^ f l> t
*g A
A
g 2 - ^ g
X
g * - ^ gg
X
~
g ' - l gVg
where
( 2
'
8 7)
33
Since Eq. (2.85) is written for a single energy group, the calculation of
the matrices [P|] and [ R | ] in multigroup problems is straightforward.
In two dimensions, [PH] and [RJ> ] contain 3 and 4 unique entries, res-
pectively, per energy group. Another important property of Eq. (2.85)
is that the source and response matrices depend only on the material
properties (specifically the diffusion coefficient and the removal cross
section) of the k-th node. Thus the unique entries of these matrices
need be computed prior to the outer iterations and stored only for
unique nodes characterized by their material composition assignment.
2.7 Boundary Conditions
The nodes are coupled by requiring the surface-averaged partial
currents to be continuous across the nodal interfaces. This is accom-
plished by using the computed outgoing partial currents as incoming
partial currents to the respective neighboring nodes, i.e.
ji n , k
( +h / 2)
_ jo u t , a
(
_
h / 2)
(2.88a)
gx gx
J
i n , A
( - h /2) = J
Ou t

k
( +h /2) , (2.88b)
gx gx
where k and I are neighboring nodes sharing the surface denoted (in
local coordinates) by x*h/2 in the k-th node and x-h/2 in the Jt-th
node. Equations (2.88) are equivalent (in an integral sense) to Eqs.
(2.3) and (2.4), i.e. Eqs. (2.88) insure continuity of the surface-
averaged values of the flux and surface-normal component of the net
current across the nodal interfaces.
The incoming partial currents on nodal surfaces which form part
of the outer boundary of the solution domain are computed in terms of
the outgoing partial current on the same surface, e.g.
J
l n , k
( +h /2) - y J
o u t , k
( +h / 2) . (2.89)
gx g gx
Recalling the general form [Eq. (2.5)] of the boundary condition,
a
g < *>
+ 2b
g "'
D
g 2 *g<
r
8
> - . (2-5)
the coefficient Y is given by
34
The standard boundary conditions are specified as follows:
-1 zero flux boundary condition
0 zero incoming partial current boundary conditions (2.91)
+1 zero net current (reflective) boundary condition.
Periodic (i.e. rotational symmetry) boundary conditions in the
hex-plane are treated by using the computed outgoing partial currents
across a boundary as an incoming partial current across the corre-
sponding periodic boundary.
2.8 Summary
The two-dimensional nodal scheme derived in this section involves
a total of 10 principal unknowns per node per group: 6 surface-averaged
outgoing partial currents and 4 spatial moments of the intra-node flux
distribution. The partial currents are calculated from the response
matrix equation, Eq. (2.85), while the flux moments are computed using
Eqs. (2.72) and (2.74). The required source moments are obtained from
Eq. (2.86). Since all terms (other than the source terms) on the right
hand side of Eqs. (2.72) and (2.74) can be evaluated in terms of the
partial currents, and the incoming partial currents are simply outgoing
partial currents from neighboring nodes, these results represent 10
equations for the 10 principal unknowns. The iterative solution of
these equations is discussed in Section 4.
35
3. DERIVATION OF THE NODAL EQUATIONS IN THREE DIMENSIONS
The three-dimensional nodal scheme derived in this section employs a
transverse integration procedure to reduce the three-dimensional diffusion
equation to four coupled one-dimensional equations, three in the hex-plane
and one in the axial direction* The hex-plane equations are approximated
as described in Section 2.4, while a more conventional cubic polynomial
approximation is applied to the axial equation. The hex-plane and axial
approximations are then combined to form a response matrix equation similar
to that derived previously in two dimensions.
3.1 The Nodal Balance Equation
The nodal balance equation is obtained by integrating the diffusion
equation over a homogeneous three-dimensional node V* defined by
V
k
: (x,y,z) xe[-h/2,+h/2], ye[-y(x),+y (x)], ze[-Az
k
/2,+Az
k
/2],
D S
where Az^ is the axial mesh spacing and, as before, h is the lattice pitch
and y
s
(x) is defined in Eq. (2.6). Recall that the origin (in local coor-
dinates) is taken as the center of the three-dimensional node. Operating
on Eq. (2.1) with
v
k
/
rev"
where V is the volume of the hexagonal-z node, and then applying
Gauss' theorem as in the two-dimensional development yields the
three-dimensional nodal balance equation
V If -If 1 -If r t -If -If
K
+ L
K
+ L
K
] + - = r - L
K
+ Z* 9 - Q^. (3.1)
gx gu gv
,
Az
k gz g *g
x
g
v
'
The node-averaged values of the flux and multigroup source term are
defined by
(3.2)
36
Az
k
/2 h/2 y (x)
I f f f>
8
"T" I dz I dx I dy
' J J J
J
- bz II '-h/2
J
- y (x)
where
Az
k
/2 h/2 y
g
(x)
V
k
= I dz I dx J dy
-Az
k
/2 -h/2 >y( x) -Az
k
/2 -h/2
( 3
-
3 )
(3.4)
*"*K ^K. ^k. ^iC
The terms L
g x
, L
g u >
L
g v
, and L
g 2
are leakages in the three hex-
plane directions (recall Fig. 5.1) and tbe axial direction, e.g.
L
k
s J
k
(+h/2) - J
k
(-h/2) (3.5)
gx gx
v
gx
%
L
k
H J
k
(+Az
k
/2) - J
k
(-Az
k
/2), (3.6)
gz gz gz
k k is
where J
gx
(h/2) and J
g Z
(Az
K
/2) are face-averaged surface-normal
components of the net current on the x- and z-directed faces:
Az
k
/2 . y (x)
-k 1 1 1 If k 3 k I
J (h/2) =1 r- I dz -s7r I dy -D ? 6 ( x, y, z ) | ..
/ o / o
,v
gx I A k I 2y (x) I ' g 3x
T
g
, /
* |xh/2 (3. 7)
* AZ I S I J
^- Az
k
/2 J (x)
k
J
k
(Az
k
/2) = - f dx f " dy -D
k
| - *"(x,y,
Z
) J
u
. (3.8)
gZ ylv I I 8 "* O
'-h/2
37
3.2 The Transverse Integration Procedure In Three Dimensions
The transverse integration procedure discussed in Section 2.3
can be used to reduce the three-dimensional diffusion equation to two
coupled equations, a two-dimensional equation in the hex-plane and a
one-dimensional equation in the axial direction. The hex-plane equation
can then be further reduced (as in Section 2.3) to three coupled one-
dimensional equations in the three hex-plane directions. However,
consistent with our approach in two dimensions, the one-dimensional
hex-plane equations are derived in a single step by first introducing
the partially-integrated quantities
Az
k
/2 y
s
(x)
dz
f
dy
*g <
x
' y ^ <
3
-
9
>
/
-Az
k
/2 "- y ( x )
9
Az
k
/2 y
s
(x)
J
g x
( x )
= I
dz
f
dy _D
g h *g
Az
k
/2 y
g
(x)
Q
k
(x) = I dz f dy Q
k
( x,y ,z) , (3.11)
and then performing a neutron balance on the slice (perpendicular
to the x-direction) defined by
6V
k
: (x,y,z) xe[x,x+ dx], ye[-y (x),+ yU)], ze[-Az
k
/2,-Wz
k
/2],
The balance equation takes the form
38
y
s
(x)
-y
s
(x)
(3.12)
where JgxyCxj-ygCx)) are z-integrated, surface-normal components of
the net current across the u- and v-directed surfaces:
Az
k
/2
J
g x y
( x
> V
x

H / dz - D
k
n V < j.
k
(x,y,z) ,
N
I . g + -
Y
g
v ,J,
y - y
e
( x )
(3.13a)
J* (x,-y (x))
gxy *
J
s
I
Az
k
/ 2
dz +D
k
n_-V (
k
(x,y,z)
O O
-Az
k
/2
y= -y
g
(x) '
(3.13b)
and the unit vectors n+ are defined in Eqs. (2.20). The final term
in Eq. (3.12) involves the axial leakage defined by
Az
k
/2
/
dz - D
k
A "" / O -Az
k
/2
k 3
zAz /2
(3.1A)
Note that
ta/2 y ( x)
_ Az
k
f . f
B
.
T
T J
y
'-h/2 '-y^x)
(3.15)
39
-k
where L
g 2
is defined in Eq. (3.6).
Equation (3.12) is very similar to the one-dimensional balance
equation [Eq. (2.18)] obtained in Section 2.3, with an additional term
accounting for leakage in the axial direction. As in Eq. (2.25), the
partially-integrated net current and flux are related by
J
g x
( x )
^ "
D
g k 4
( X > + D
8
y
s
where
(
L v
( x
' y
)
f
dz
t
gxy I g
The one-dimensional axial equation can be obtained by either
operating on the three-dimensional diffusion equation with
h/2 y (x)
s
/ - J
dy ,
-h/2 -y
e
(x)
5
or by performing a neutron balance on
6V
k
: (x,y,z) xe[-h/2,-tti/2], ye[-y
e
(x),-l7
o
(x)], ze[z,z + dz],
The result is
z
T
where, analogous to Eqs. (3*9) - (3.11),
40
h/2 y
g
(x) /
I d x J
-h/2 '-
J dy < ^(x,y,z) (3.19)
-h/2 '
s
h/2 y
s
(x)
^-h/2 ^-y
s
(x)
h/2 y
s
(x)
'-h/2
J
-yU)
dy - D J|J. *
k
(x,y,z) (3.20)
dy Q^(x,y,z). (3.21)
The partially-integrated hex-plane leakage Lg
X
y(z) is given by
y
e
(x)
> s r 2 2 ~\
L
Ly
( z ) E
I
dx
/
dy
"
D
e r ~ 2
+ i
T Ut
(x
'y
z
>- (
3
-
22
>
/I - /:
-h/2 '-y (x)
The total hex-plane leakage is
r
_ 3h 1 I
V J
L
,
" =1 ^- V I dzL
k
(z), (3.23)
gxy 2 ,,k / . gxy
v
-Az /2
which can be written in terms of the average leakages in the three
hex-plane directions,
If V If W
K
= L
K
+ L
K
+ L
K
. (3.24)
gxy gx gu gv
We also introduce the hex-plane averaged quantities
41
h/2 y (x)
s
* / . - /
(3.25)
-h/2 " - y (x)
s
h/2 y
s
(x)
, dx / dy - D ^- ^ ^(x, y , z). (3.26) I
-h/2
Since Eq. (3.18) is in P-l form, it is necessary (once again) to
specify a relationship between the partially-integrated net current and
flux. However, since the integration limits in Eq. (3.19) do not depend
upon z, this relationship between JH
z
(z) and <j>^
2
(z) is particularly
simple:
3
g h 4
( z )
*
(3
-
27)
Similarly,
3
g d 7 *g z
( z )
-
( 3
-
28)
The one-dimensional hex-plane equations [Eqs* (3.12) and (3.16)]
are approximated as described in Section 2*4. Thus we now turn to the
development of an approximation scheme for the one-dimensional axial .
equations [Eqs. (3.18) and (3.27)].
42
3.3 Approximation of the One-Dimensional Axial Equations
3.3.1 The One-Dimensional Axial Polynomial Approximation
The one-dimensional axial flux is approximated as follows:
N
5
*
l
*
n=1
( 3
*
29 )
where
k
. = *
k
(+Az
k
/2) - f
k
(-Az
k
/2)
gzl gz gz
(3.30a)
a
k
= ? (+Az
k
/2) +
k
(-Az
k
/2) - 2^
k
gz2
T
g z gz g
(3.30b)
f
z l
( z) H -^- =
Az
(3.31a)
f
z2
(z) = 35
2
- (3.31b)
f
z 3
( z ) = (3.31c)
As in the hex-plane approximation, the axial approximation is con-
structed, such that the node-averaged flux <j>g and the surface-averaged
fluxes < j>
ez
(Azk/2) are preserved, i.e.
V
k
/
Az
k
/ 2
dz (3.32a)
-Az
k
/2
43
Consistent with Eqs. (3.32), the basis functions defined in Eqs. (3.31)
satisfy the constraints
J
Az
k
/2
dz f (z) i 0, n = 1,2,3, (3.33a)
zn
-Az
k
/2
f
z3
(Az
k
/2) = 0. (3.33b)
Comparison of Eqs. (3.16) and (3.27) shows that unlike the one-
dimensional hex-plane fluxes, the axial flux <j>|
z
(z) does not exhibit
a first-derivative discontinuity. Thus the basis function f
z
3(z) is a
simple cubic polynomial subject only to the constraints given in Eqs.
(3.33).
Equation (3.29) is equivalent to the polynomial used to approximate
the one-dimensional fluxes in the Cartesian-geometry nodal expansion
method
15
(NEM). The NEM formulation however differs from the Cartesian-
geometry analog of the method developed here in that the NEM equations
are not cast in the multidimensional response matrix form shown in Eq.
(2.85). Although NEM calculations for light water reactors (LWR)
typically use a fourth-order polynomial expansion [i.e. N
z
= 4 in Eq.
(3.29)], only a third-order (cubic) polynomial is used to approximate
the one-dimensional axial flux here. This choice is based on the
observation that while LWR nodes are often 10 diffusion lengths in
dimension, the 15 to 20 cm axial burnup regions typical of fast reactor
calculations are equivalent to axial node dimensions of only 3 to 4
diffusion lengths.
3.3.2 Calculation of the Expansion Coefficient
The expansion coefficient aj$
z
3 is calculated by applying a
weighted residual approximation to the axial balance equation, Eq. (3.18).
As in Section 2.4.4, we weight the one-dimensional balance equation with
weight function w (z), and then require the result to be zero when
integrated over the interval ze[-Az
k
/2,+Az
k
/2], i.
e
.
44
where the inner product is defined by
Az
k
/2
<w(z), <j>
k
(z)> H i - f dz w(z) <|>
k
( z) .
*^- Az
k
/2
As before, it can be shown that unit weighting, i< e.
w(z) = w
n
(z) = 1,
reduces Eq. (3.34) to the three-dimensional nodal balance equation,
Eq. (3.1).
Several choices of weight functions are possible for the calculation
of a|
z
3. These choices include "half-node" weighting such as that used
in tne hex-plane approximation, i.e.
w(z) = w
z l
( z ) = sgn(z), (3.35a)
moments weighting,
w(z) = w
z l
( z ) = - ^ E f
z l
( z ) , (3.35b)
Az
and Galerkin weighting,
w(z) =w
?1
(z) =f
z 3
( z) . (3.35c)
Numerical comparisons
15
using the nodal expansion method have shown
moments weighting to be more accurate than Galerkin weighting. Similar
studies
13
during the early stages of the present work support this con-
clusion. Additional slab-geometry results have further demonstrated that
moments weighting is slightly more accurate than half-node weighting.
Thus Eq. (3.34), with weight function w
z
j(z) defined in Eq. (3.35b), is
used to calculate aj|
z
3.
The axial spatial moment is thus defined by
*g zl
H < W
z l
( z
> '
V /
Az
k
/2
k
Az
K
/2
(3.36a)
Az
k
/ 2 h/ 2 y (x)
f Z I K
~Az
k
/ 2 h/ 2 y
o
( x)
S
Substituting Eq. (3.29) into Eq. (3.36a) and performing the necessary
integrations yields
1 ft A K
120
l l O a
gzl
or, using Eq. (3.30a),
a
k
, 120 <|>
k
. + 10 [
k
(+Az
k
/2)
k
(Az
k
/2)] . (3.37)
gz3 gzl
T
gz gz
The threedimensional nodal scheme thus involves a total of five flux
moments: the nodeaveraged flux, three hexplane moments, and one
axial moment. Equations for these moments are derived in the following
subsection.
3.4 The Flux Moments Equations
Th e zeromoment or nodeaverag ed flux satisfies th e nodal balance
equation [Eq. (3.1)]:
k 1 k 2 k k k 1 k
AT ^rr (t = r [IT + L* + L
K
] . . L
K
. (3. 38)
' g
E
r , k
v
g
3 M
r , k gx gu gv
J
Az
k
E
r , k gz
3 M A z E
g g g
46
This result can also be obtained via unit weighting of either Eq. (3.12)
or (3.18).
The hex-plane moments equations are derived from Eq. (3.12) using
the same procedure as in Section 2.5. The x-direction result is
D
k
f
k k
.. ._ T* - T* ] +:- & a * ,. (3.39)
Jh gx gu gv
1
9h h gxl
This equation is very similar to Eq. (2.70). The additional term
h/2 y_(x)
T
k _ Az
Li
f f
s
*
I dx sgn(x) I dy L (x,y)
*
/
-h/2
J
~ y (x)
h/2 y (x) Az
k
/2
s
k f f f
2
= f- I dx sgn(x) I dy I dz - D
k
5_^ <(,
k
(x,y,
Z
) (3.40)
-h/2 "-y (x)
s
is the x-direction moment of the partially-integrated axial leakage.
The axial flux moment equation is derived from Eq. (3.34) with
w(z) given by Eq. (3.35b):
<
w
zl<
2
> k
J
g z
< Z ) > +
'g
,
" *g
Z
l i *l - 3h
L
gxyzl>
where the axial flux moment <j> , is defined in Eqs. (3.36), and
Az
k
/2
-Az
k
/2
< >
47
Az
k
/2 h/2 y
e
(x)
2
r f dz-2r- f dx f dy -D
k
U- y + i _ Lj(
Xf
y
i 8
) (3.42b)
f Az
K
I I
g
ldx 3y J
8
is the axial moment of the partially-integrated hex-plane leakage.
Integrating the first term in Eq. (3.41) by parts yields
' dT
J
g
Z
(z)>
- ~
J
k k V (3.A3)
A z
K K g Z
where Jg
Z
(z) and a
g z
i are defined in Eqs. (3.26) and (3.30a),
respectively. Substitution of Eq. (3.43) into Eq. (3.41) yields
after rearrangement
rfk
=
J L _ r n
k
'gzl
E
r,k
l
V
2
" 3h
L
k
gxy
D
k
Az
k
z l
]
Az
1
1
2
A
r w if
g
1
k
g!
''g
-f
k
t gz
where
(3
'
4 4
>
Equation (3.39) and its u- and v-direction analogs can be combined
with Eq. (3.44) to yield
48
P
gxl
k
P
g u l
.k
''gvl
.k
P
g zl
h
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 a
L
Az
k 8 Z u l
gzvl
Q - = L
""*' 3h gxyz 1
where
3a
gl
1 1 -1
1 1"
-1 1 0
0 0 0 a

k
g3-
gx
T
k
gu
T
k
gv
T
k
L gz J
9a
k h
0 0
0 0
0 0 0 a "
a
gul
k
gvl
k
(3.46)
a
k _
g
(3.47a)
49
3 ,
V = k r k
(3
'
47b>
8
4Az 2
g
9ot
k
^
( 3
'
4 7 c )
^T k r k
Az
k
Az
R
E
r

k
and otgi is defined in Eq. (2.73). The calculation of the leakage
moments L g
Z X
i , Lgzul*
L
gzvl
an<
^
L
gxyzl *
s
discussed in Section 3.6.
3.5 The Response Matrix Equation
The face-averaged partial currents in three dimensions are
defined by
k
/2 _ y (x)
i f
dz
[lyfuT f
S
< * 1| JCx.y..)
*'-Az
k
/2 '-y (x)
s
Az
J . k .. J
- y
Az
k
/2 y (x)
T i n . k , ^ .
/ ON
_ 1 T j f l C j j l - i k i
J-. .
,
(ih/Z) = j - / dz I ..
/
. .
x
/ dy {
T
^<
Az
K
-Az
k
/2
* " ^ 1/ "K W l A t V * 7 I I . . *
50
h/2 y
s
(x)
, k
/ + A
k
/ O N
_ A z I , I , I 1
( z /2) =
"k ~ I
dx
/
dy
| i
'-h/2 *'-y
c
(x)
k
( 3
'
4 9 a )
zAz
k
/2
h/2 y
s
(x)
7in,k
/+A
k
J
g z
(Az
*'-h/2 '-y (x)
i D^ !- Nx .y .o ] . . (3.49b)
2
8
3 z
8 JzAz
k
/2
The surface-averaged fluxes and net currents can be written In terms
of the face-averaged partial currents using the relationships shown
in Eqs. (2.76) and (2.77).
The equation used to compute J ' (+h/2) is identical to
Eq. (2.80) derived in Section 2.6:
g x
D
jout.k . _^
f
k 36 k 7 k 1 k k
f
j
gx h gxl 13 gx2 26 gx3 2 gx4 gx
v
+ J
l n

k
(+h /2). (3.50)
gx
gx
A similar equation for Jgg ' (+Az
k
/2) is derived by using the one-
dimensional axial polynomial [Eq. (3.29)] to evaluate the derivative
in Eq. (3.28). Substituting this result into the z-direction analog
of Eq. (2.77a) yields
D
k
* + 3.J
+ a ] +
J^
k
(^^
k
/2) (3.51)
51
The derivation of the final form of the three-dimensional response
matrix equation follows the five steps described in Section 2.6. We
repeat these steps here, with some obvious overlap, for the three-
dimensional case. Additional details are provided in Appendix B*
k k k k
(1) The expansion coefficients a
g
xl>
a
gx2
a
gx3
an(
*
a
g x4
a r e
are eliminated from Eq. (3.51) using Eqs. (2.34a), (2.34b), 12.53) and
(2.59b), respectively. The term E|(h/2) is eliminated using Eqs. (2.43)
and (2.44). The expansion coefficients a &
z
i , a^
z
2, and a|
z
3 are
eliminated from Eq. (3.53) using Eqs. (3.30a), T3.30b), and (3.37),
respectively.
k "4c k
(2) The flux moments <J>
g
, <j>g
X
l> and 4>gzl introduced via step,
(1) are eliminated in favor of the source moments Q
g
, Qg
x
l> and Qgzl
and the leakage moments LJ
zx
i and Lgxyzl using Eqs. (3.38), (3.39),
and (3.44), respectively.
(3) All surface-averaged fluxes and net currents introduced via
steps (1) and (2) are eliminated in favor of surface-averaged partial
currents using Eqs. (2.76) and (2.77) and their z-directed analogs.
Equations (3.50) and (3.51) can then be written as
r
t a
l
a
2
a
3
a
4
a
3
a
2
a
5
T
out,k
-g
b
2
]
_
L
k
z
k gzxl,
c
5
] J
g
in,k
(3.52)
and
[a
6
a
6
a
6
a
6
a
6
a
6
a
7
a
8
]
Q
k _ 2_
L
k
gzl 3h gxyzl
e
[c c
?
c
8
]

(3.53)
respectively, where J
o u
' and J
n
* are column vectors containing
the eight outgoing ana eight incoming partial currents, respectively,
for the k-th node, e.g.
52
k
i col ( J f ( - H , / 2 ) , J ^ N + h / a ) , J (
+
h / 2 ) , J
g gx gu gv gx
gu ' gv gz 8
Z
The constants a^, b^, and c^ are defined in Appendix B.
(A) Five additional equations similar to Eq. (3.52) can be
obtained by applying successive 60 rotational transformations to Eq.
(3.52). Calculation of J2
, k
( - Az
k
/ 2) yields an additional equation
similar to Eq. ( 3.55) . These additional equations can be combined
with Eqs. (3.52) and (3.53) to yield
[A
k
] j
o u t
'
k
- [
B
k
] {Q
k
- L
k
} + [C
k
] J
i n, k
, (3.55)
where
Q
k
i col [Q
k
, Q
k
., Q
k
., Q
k
,, Q
k
J (3.56)
g 8 gxl*
x
g u l '
H
g v l ' ^gzl
J
and
L
k
= col [0, -~ L
k
, , - VL
k
, , - VL
k
,,|7-L
k
J. (3.57)
-8
A z
k
gzxl
,
Az
k gzul
,
Az
k gzvl' 3h gxyzl
J
(5) The final form of the response matrix equation is obtained
by inverting [A
k
] in Eq. (3.55) to yield
j k _ jpkj
{ 2
k _ jkj +
[ R
k
] j
i n
'
k
, (3.58)
o o o o o o
where [P
g
] is an 8 by 5 source matrix and [R
g
] is an 8 by 8 response
matrix equation.
As before, the source term is calculated using
I
x
g 2-r
V
V h'
+
2 j
E
gg' <
(3
--
59)
53
where
k
= c o l fo
k
, $
k
,, < ]>
k
., < |>
k
., |.
k
. ] . ( 3 .60 )
8 8 g x l
,
*g u l *
y
g v l '
y
g z l '
k k
In three dimensions) [Pg] and [R
g
] contain 5 and 8 unique
entries, respectively* As in two dimensions, these entries depend
only on the material properties and axial mesh spacing of the k-th
node, and thus need be computed and stored only for unique nodes
characterized by their material composition assignment and axial
mesh spacing*
The incorporation of boundary conditions into the global
solution of Eq. (3.58) is accomplished in the same manner as
discussed in Section 2.7.
3.6 Calculation of the Leakage Moments
The x-direction moment [Eq. (3.40)] of the partially-integrated
axial leakage [Eq. (3.14)] is calculated using the approximation
L
k
z
(x, y ) = L
k
z
. (3.61)
Thus the space-dependent axial leakage is simply replaced by
its average value over the z-directed faces. Substitution of Eq.
(3.61) into Eq. (3.40) and its u - and v-direction analogs yields
L
k
. = L
k
. = L
k
. i 0. (3.62)
gzxl gzul gzvl
The axial moment [Eq. (3.42)] of the partially-integrated
hex-plane leakage [Eq. 3.22] is calculated using the approximation
P ( Z > ZE A
*
k

(3
'
6 3 )
where P
g X
y (z) is a quadratic polynomial:
p
K
(z) = L
K
+ p
K
. f .(z) + P
0
gxy gxy gxyl zl gxy2
54
Here f
z
j(z) and f
z
2(z) are as given in Eqs. (3.31a) and (3.31b), and z\
and Z2 will be defined shortly* This "quadratic leakage" approximation
15
has been used extensively in the development of recent Cartesian-geometry
nodal schemes.
1 5
"
20
Setting Pg
X
yl = p|xy2 =
i n E<
1* (
3
64) reduces
Eq. (2.63) to a "flat leakage" approximation as in Eq. (3.61). The
coefficients Pg
X
yl and Pgxy2 are calculated in the following manner.
Let k- and k+ denote the neighboring nodes in the minus and plus
z-directions, i.e. the nodes immediately below and above the k-th
node, respectively. The coordinates z\ and Z2 are defined by
z, = -Az
k
/2 - A z
k
~ (3.65a)
= +Az
k
/2 + A z
k+
, (3.65b)
and thus the quadratic polynomial extends over the three nodes k-, k,
and k+. The expansion coefficients in Eq. (3.64) are calculated such
that the total hex-plane leakages [Eq. (3.23)] in the nodes k- and k+
are preserved:
-Az
k
/2
L
k
~ =?r-r- I dz p
k
(z) (3.66a)
gxy 2 k- I gxy
v
-k+ _ 3h _1 _ f
gxy
=
2 yk+ /
Z
2
dz P
k
(z). (3.66b)
+Az
k
/2
The required leakage moment is calculated by substituting Eq. (3.63)
into Eq. (3.42a) and performing the necessary integrations:
k 3h Az^JL k
gxyzl * 2
v
k 12 "
As shown in Section C.I of Appendix C, this procedure leads to the
final result
55
L
k
. = u
k
[L
k
- L
k
~ ] + u
k
[L
k+
- L
k
], (3.68)
gxyzl - gxy gxy
J
+ gxy g xy
J,
k k
where the coefficients \i _ and u depend only upon the axial mesh
spacings:
k
y
k
= ^ r [2Az
k+
+ A z
k
] [A z
k+
+ A z
k
] (3.69a)
A z
k
] [Az
k
" + A z
k
] (3.69b)
d = [Az
k_
+ A z
k
] [Az
k
~ + A z
k
+ A z
k+
] [Az
k
+ A z
k +
]. (3.69c)
The incorporation of axial boundary conditions into the calculation
of these coefficients is discussed in Appendix C. Using Eq. (3.24),
the total hex-plane leakages shown in Eq* (3.68)_can be computed in
terms of the directional leakages L
g x
, L g
u
, and Lg
v
which, in turn,
are readily calculated using the available face-averaged partial
currents.
The quadratic approximation to L g
x
y (
z
) is necessary in order to
obtain sufficient accuracy using coarse (~20 cm) axial meshes. The
flat approximation to LJ|
z
(x,y) is clearly less accurate, although the
error due to this approximation will remain within acceptable limits
provided the second derivatives of LJ
z
(x,y) over the node are (in
some sense) small. This should be the case in typical LMFBR designs
in which the lattice pitch is ~11 to 16 cm, or only 2 to 3 diffusion
lengths.
The contribution to the total error due to the approximations
introduced in Eqs. (3.61) and (3.63) is analyzed in Appendix C for a
simple homogeneous model problem. The results of this study confirm
the accuracy of the quadratic approximation to L ^
X
y (z), and further
suggest that the error due to the flat approximation of Lo
Z
(x,y)
represents a significant contribution to the total error. It is
clear that this error could be essentially eliminated by introducing
a quadratic approximation analogous to Eq. (3.63), i.e.
dy L
k
z
(x, y) 2 p
k
z
(x), (3.70)
y
e
(x)
56
where p
g Z
(x) is a quadratic polynomial extending over the two neighboring
nodes in the x-direction. Use of this approximation plus the analogous
expressions in the u- and v-dlrections leads to expressions of the form
shown in Eq. (3.68), only involving axial leakages in the six neighboring
nodes in the hex-plane* From a computational viewpoint, the calculation of
the axial moments using Bq. (3.68) is relatively straightforward; however,
the additional hex-plane coupling introduced by Eq. (3*70) would greatly
complicate the solution of the partial currents. Furthermore, although the
results in Appendix C suggest that the flat leakage approximation has a
significant effect on the overall accuracy of the nodal scheme, the numerical
results presented in Section 5 demonstrate that in spite of Eq. (3.61), very
acceptable accuracy is obtained in three-dimensional nodal calculations. For
these reasons, the approximation given in Eq. (3.70) has not been implemented
in the DIF3D nodal option.
3.7 Summary
The three-dimensional nodal scheme derived in this section involves
a total of 13 principal unknowns per node per group: 8 surface-averaged
outgoing partial currents and 5 spatial moments of the intra-node flux
distribution. The partial currents are calculated from the response matrix
equation, Eq. (3.58), while the flux moments are computed using Eqs. (3.38)
and (3.46). The required source moments are obtained from Eq. (3.59), and
the leakage moments are calculated using Eqs. (3.62) and (3.68).
57
4. NUMERICAL SOLUTION OF THE NODAL EQUATIONS
4.1 Overview of the Solution Procedure
The nodal equations are solved using a conventional fission source
iteration procedure
21
accelerated by coarse-mesh rebalance
21
'
22
and
asymptotic source extrapolation.
23
At each fission source (or "outer")
iteration, the interface partial currents for each group are computed by
solving the response matrix equations with a known group source term.
This solution is accomplished via a series of sweeps through the spatial
mesh. These sweeps, which are discussed in Section 4*2, are analogous
to the "inner" iterations used to invert the in-group diffusion-removal
matrix in the finite difference option. The coarse-mesh rebalance and
asymptotic source extrapolation procedures are described in Sections 4*3
and 4.4, respectively.
The algorithm used to solve the nodal equations is shown in Fig.
4.1. The nodal coupling coefficients are computed prior to the start of
the outer iterations (n is the outer iteration index). The solution
vectors (i.e. the flux moments, interface partial currents, and fission
source moments) are initialized by assuming a spatially constant flux
distribution in each energy group. As shown in Fig. 4.1, the loop over
energy groups performed at each outer iteration consists of the following
steps:
(1) The coarse-mesh rebalance factors and asymptotic source extra-
polation factor computed at the previous outer iteration are applied to
the partial currents and fission source moments as described in Sections
4.3 and 4.4.
(2) The group source term of outer iteration n is computed using
Eq. (3.59):
rt
k(n) 1 , k(nl) , \ *.s,k ,k(n)
8'<8
k
where ^ is a vector containing the fission source moments, i.e.
g'-i
58
I C O MP U T E NODAL COUPLING COEFFICIENTS
I
I N I T I AL I Z E S O L U T I O N VECTORS
(O U T E R I T E R AT I O N LOOP)
(L O O P OVER GROUPS)
APPLY AC C E L E R AT I O N FACTORS TO SOLUTION VECTORS
CONSTRUCT GROUP SOURCE TERMS
DUE TO F I S S I O N AND I N -S C AT T E R
1
SOLVE RESPONSE MATRIX EQUATION FOR P AR T I AL CURRENTS
(
| COMPUTE FLUX MOMENTS
1
COMPUTE CROUP CONTRIBUTION TO NEW FISSION
SOURCE MOMENTS AND C O AR S E -M E S H REACTION RATES
YES
NO
S O L V E C O AR S E -M E S H R E B AL AN C E EQUATIONS
FOR R E B AL AN C E FACTORS AND E I G E N V AL U E
CHECK CONVERGENCE OF F I S S I O N SOURCE
NO I G E N V AL U E AND FISSION
SOURCE CONVERGED ?
I SOLUTION E D I T |
C O NTR O LLI NG
SUBROUTINE
D NHCCC
O NH S T T
OUTR1
0UTR2
O U TR 3
0U T R 4
O U TR 4
O U T R 5
OUTRS
O NH F I N
Fig. 4.1 Overview of the N odal Solution Algorithm
59
and *
n
' is the eigenvalue computed at the previous outer iteration.
The in-scatter contribution in Eq. (4.1) is evaluated assuming no
up-scatter.*
(3) The interface partial currents are computed by solving the
response matrix equations [Eq. (2.85) or Eq. (3.58)] with known source
terms Q*p
n
' This calculation involves a series of inner iterations
discussed in Section 4.2.
(4) The flux moments are calculated from either Eqs. (2.72) and
(2.74) (two dimensions) or Eqs. (3.38) and (3.46) (three-dimensions)
using the most recently computed partial currents to evaluate all terms
in these equations with the exception of the transverse leakage moment
Lgxyzl *
n
Eq. ( 3.46) . This latter term retains its value used in the
calculation of the partial currents at the final inner iteration
preceeding the flux-moment calculation.
(5) The calculated flux moments are then used to compute the
group contribution to the new fission source moments [Eq. ( 4.2) ] . The
group contr'Vttions to the reaction rates and leakages required for the
coarse-mesh ance equations are also computed.
Once all ^nergy groups have been processed, the coarse-mesh
rebalance equations are solved and the convergence of the fission source
is checked for asymptotic behavior. The outer iterations are terminated
when the following convergence criteria are satisfied:
Eigenvalue: - A
(n-1)
(4.3a)
Pointwise Fission Source: max
. J
c(n-l)
(4.3b)
Average Fission Source: T7
1/2
(4.3c)
The default values of the convergence criteria e , e
2
1.0 x 1 0 ~
7
, 1.0 x 1 0 "
5
, and 1.0 x 1(T
5
, respectively.
and e, are
Up-scattering is not permitted in the DIF3D nodal option.
60
4.2 Solution of the Response Matrix Equations
4.2.1 Two Dimensions
The two-dimensional response matrix equations are solved in each
group at each outer iteration by sweeping the nodes in the "four color
checkerboard" ordering shown in Fig. 4.2. This ordering is motivated by
the "red-black checkerboard" ordering (often referred to as aj - ordering
21
)
in Cartesian geometry. Note that the hexagons are assigned colors
(i=1,...,4) such that two hexagons of the same color do not share a
common surface. (It is also possible to color a hexagonal map using
only three colors.) The ordering in Fig. 4.2 suggests an iterative
procedure based on a mesh sweep (or inner iteration) consisting of four
passes through the mesh in which all outgoing partial currents from nodes
of color i are computed during the i-th pass. Letting n and ml denote
the outer and inner iteration indices, respectively, the local response
matrix equation solved at each node is
jout,k(n,ml)
= fp
k,
Q
k(n )
+ fR
k,
J
in,k(n,ml/ml-l) .^
-g g *g g -g *
where the incoming partial currents are the most recently computed out-
going partial currents from neighboring nodes. The incoming partial
currents on the outer boundary are updated at the end of each inner
iteration using Eq. (2.89). Models with either sixth- or third-core
symmetry are solved by sweeping only over those nodes contained in the
fractional-core region of solution.
Fig. 4.2 The Four-Color Checkerboard Ordering in Hexagonal Geometry
61
The choice of this inner iteration procedure is based on two primary
considerations: First, because the convergence rate of the outer iterations
can be reduced by the introduction of higher harmonics during the inner
iterations, it is important that the partial currents be computed such that
any symmetry inherent in the problem be preserved by the inner iteration
procedure* For example, given a third-core symmetric source distribution,
a single mesh sweep in the ordering shown in Fig. 4.2 will produce a partial
current solution rhich is exactly third-core symmetric* Although this scheme
does not preserve exactly either sixth- or twelfth-core symmetry, it does
so approximately since exact twelfth-core symmetry would be preserved for
1H2 and 324 in Fig. 4.2.
The second consideration in the choice of an iterative procedure is
the following: Since the structure [see Eq. (A.49)] of [R|] permits very
efficient coding of the operations necessary to calculate simultaneously all
outgoing partial currents from a node, it is considered important that Eq.
(4.4) be solved in such a manner. For example, although a scheme in which
all hexagons on a ring are solved simultaneously preserves all possible
symmetries, it sacrifices the computational advantages inherent in Eq. (4.4)
by requiring the formation and solution of penta-diagonal matrix equations
for the partial currents across surfaces shared by hexagons on the same ring.
Other schemes which are contrary to one or both of the above considerations
include (a) solving simultaneously for all x-directed partial currents on
each it-line, followed by similar solutions on u- and v-lines, and (b) solving
simultaneously for all outgoing partial currents from all nodes on an x-line.
This latter scheme is analogous to the line over-relaxation procedure
employed in the DIF3D finite difference option. Thus, in conclusion,
the four-color checkerboard sweep appears to offer a compromise between
preserving inherent problem symmetry and exploiting the computational
advantages associated with the direct solution of Eq. (4.4) for each node.
This inner iteration procedure is equivalent to a Gauss-Seidel
iteration applied to the global response matrix equation.* To
demonstrate, we introduce the global partial current vector
T
out . r.out .out
T
out -out
T
in,B, .. ,.
-gxy -gxyl' -gxy2* -gxy3* -gxy4* -gxy '
where
J
o u
. contains all outgoing partial currents for nodes of
"
g x y i
color i, i - 1,...,4,
*In the following discussion it is assumed that the outgoing partial currents
from all nodes of the same color can be solved simultaneously, i.e. that any
node can be decoupled from all other nodes of the same color. This is
rigorously true only for the full-core model shown in Fig. 4.2* It is not
true for fractional-core models (see Fig. 7.2) with periodic boundary con-
ditions since nodes of the same color may be coupled via the periodic boundary
conditions* The terms introduced by this periodic coupling have been omitted
in the development of the two-dimensional Gauss-Seidel procedure in order to
simplify the presentation*
62
and
J TJ
J ' contains all incoming partial currents for group g on the
~
g x y
outer boundary.
The global source vector is
S = col [S ., S
o
, S ,, S ,, 0] ,
gxy gxyl' gxy2' gxy3' gxy4
, J , (4.6)
where
k k
S contains the terms [P ]Q for all nodes of color i.
gxyi g
a
8
Eliminating incoming partial currents across the interior surfaces
in favor of outgoing partial currents, and then combining the local
response matrix equations for all nodes with the boundary conditions
[Eq. (2.89)] yields the global response matrix equation for group g:
[R
x y
] j
o u t
<
n
> .
S
( n )
.
g gxy gxy
(4.7)
The global) response matrix is
[I]
"
[ r
21
"L
12
J
[I]
I r
32"
lr
42
J

Lr
B2
J
"
[ r
13
J
"
t r
23
J
[I]
t'
43
J
"
[r
B3
J
"
tr
24
J
~
[r
34
J
[I]
[ r
2B
]
[
r
3B
]
[I]
(4.8)
where
if
[r..] and [r
1B
l contain entries of [R ] for nodes of color i,
63
[r_.] contain th e boundary coefficients Introduced In E. (2 *89),
and [I] is th e identity matrix * Applying th e GaussSeidel procedure to
Eq. (A.7) yields
T
out(ml) _(n) , ,
T
out(ml1) , ,
T
out(ml1)
S
g x yl
+ [ r
1 2
]
*g
+ t r ]
^
+ I
r
,J J
0
"'!"
1
"
0
* [r
lR
] J
1
"'
14
J
gxy4 IB gxy
T
out(ml)
c
( n)
r
,
T
out(ml) , , out(ml)
^
+ l r
41
J
^ g x yl
+ l r
42
J
ig
+ [r.J J
o u t
(
m l
) +
[ r
] jin.B(mll)
43 g x y3 46 g x y
in,b(ml) _
r
,
T
out(ml) ,
f
,
T
out(ml)
r
,
T
out(ral)
g x y
_ l r
Bl
J
^ g x yl
+ U
B 2
J
^ g x y2
+ l r
B3
J
^
m l = 1
>
M1
g
<
4

9
>
Th e outer iteration index n h as been dropped from th e partial current
vectors to simplify th e notation.
Th e number (Mlg ) of inner iterations per outer iteration in g roup
g is determined in th e following manner. Let
1
kl
wh ere th e summation is over all nodes in th e reactor. Th us *g TT is simply
th e reactoraverag ed value of th e node dimension (i.e. th e lattice pitch )
measured in diffusion leng th s.* Th e converg ence rate of th e iterative
*Th e diffusion leng th L
g
is defined by L^ = D
g
/ I
r
,
64
procedure shown in Eq. (4.9) increases with increasing tc
g
h since the
spectral radius
21
of the Gauss-Seidel iteration matrix decreases with
increasing node size* The decreased spectral radius of the iteration
matrix is due to the decreasing values of the transmission coefficient
(see Appendix A ) with increased node size, which in turn increases the
diagonal dominance of the global response iw:i.rix [R
g
]. In view of
this observation, plus numerical results for a number of test problems,
the following simple formula is used to determine the number of inner
iterations to be performed at each outer iteration:
K h > 1
g
M l
g
E I . (4.11)
< ~T < 1
In only one problem studied to date has a value of K_h < 1 been
observed. (This occurred in group 2 of the SNR benchmark problem dis-
cussed in Section 5.2.) Thus, in two-dimensional calculations, only
two Gauss-Seidel iterations of Eq. (4.7) are typically performed in
each energy group at each outer iteration.
4.2.2 Three Dimensions
Before discussing the solution procedure in three dimensions,
it is convenient to partition the local three-dimensional response
matrix equation into two coupled equations for the hex-plane and axial
partial currents:
k
= [p
k j k
+ [R
k j jin.k
+ [R
k j jin.k (4.12a)
-gxy gxy
J
*g gxy -gxy gxyz -gz
out.k
m
k j i k _
L
k
}
j k j in,k
f
k in,k
-gz
l
g z
J l
*g -g
J
gz
J
-gz gzxy
J
-gxy '
v
'
where J
O u
* and J
o u
* contain, respectively, the six outgoing
partial currents in the hex-plane and the two outgoing partial currents
in the z-direction for the k-th node. The structures of the sub-matrices
introduced here are shown in Eqs. (B.30) and (B.31). Using the leakage
approximations shown in Eq. (3.62), the leakage-moment vector defined
in Eq. (3.57) becomes
[0, 0, 0, 0,
65
As shown in Eq. (4.12a), the leakage moment shown here does not enter
into the calculation of the hex-plane partial currents*
Equations (4.12) are solved by sweeping the axial mesh planes in a
standard red-black checkerboard ordering, i.e. the odd-numbered planes
are processed during the first pass, followed by the even-numbered planes
on the second pass. The following calculations are performed for each
plane encountered in this two-pass axial sweep: (1) The hex-plane
partial currents are calculated using Eq. (4.12a) as the nodes are swept
in the four-color checkerboard ordering described in the previous sub-
section. (2) These hex-plane partial currents are used in conjunction
with the corresponding hex-plane partial currents on the two neighboring
planes to compute the leakage moment Lg
X
yzl
a s
shown in Eq. (3.68).
(3) The outgoing z-directed partial currents are then computed using
Eq. (4.12b) during a single (sequential) sweep of the nodes on the
plane. The incoming partial currents on the axial boundaries are com-
puted as in Eq. (2.89). Two complete sweeps in the axial direction are
performed in each group at each outer iteration.
As in two dimensions, this iterative procedure is equivalent to
a Gauss-Seidel iteration of a global matrix equation. Let
T
out _
n r T
out, 1
T
out, 3 ,
J ,, = col [J , J , ... J
-gxy,odd "gxy ' -gxy
,out _ ,
r
.out, 2
T
out, 4 ,
J = col [J , J * , . . . ] ,
-gxy,evn -gxy -gxy
where J^
t ,
denotes J
0
^ [defined in Eq. (4.5)] written for the A-th
plane. Thus J
u
,, and J contain all hex-plane partial currei
for the odd- and even-numbered planes, respectively. Furthermore, let
J
o u
j, and J
o u
contain outgoing z-directed partial currents
-gz,oad -gz,evn
for all nodes on odd-and even-numbered planes, respectively,
J ' contain all incoming z-directed partial currents on the
8
outer axial boundaries,
L ,, and L contain the leakage moments L . [see Eqs.
(3.42; and (3.68)] for all nodes on odd- and even-numbered
planes, respectively,
and
66
a n d
^gxyevn
c o n t a i n t h e t e r m S [P
gxy
l2
g
f o r a 1 1 n o d e s
on odd and evennumbered planes, respectively,
k k
S and S contain the terms [P ]Q for all nodes
gz.odd gz,evn gz *g
on odd and evennumbered planes, respectively.
The global matrix equation is
[r
z x y
[0]
[0]
U
z
2 1
]
[0]
to]
] [I]
[0]
[rf
2
]
[ r
2
]
[0]
~
t r
B l
]
[0]
[I]
[0]
[0]
[0]
[0]
[0]
[0]
tM
Z
2
]
[r
xy
]
[r
z x y
]
W
z
n
]
[0]
[rf
Z
]
lr
Z
]
[0]
[0]
[I]
[0]
r Z ,

[ r
B 2
]
[0]
[0]
[0]
[0]
l ?
Z
2
)
[I]
[0]
fr
x y Z
] ir
lB
J
[r
Z
B
]
[0]
[r
x y z
]
r Z !

[ r
2B
]
[0]
[I]
jOUt
gxy,odd
T
out
gz,odd
g,odd
,out
gxy,evn
.out
gz,evn
L
g,evn
gz

"
s
(n)
gxy,odd
s
(n)
gz,odd
Q
s
(n)
gxy,evn
s
(n)
gz,evn
0
0
(4.13)

where, with reference to Eqs. (4.12),

[r.
y
], 1 1,2, contain entries of [R ] for nodes on odd and
i 8^y
even numbered planes, respectively,
. "] and I *' ] , i 1,2, contain entries of [R ]
i ID gxyz
lr
y
J, i 1,2, contain entries of [R ] ,
i ' gzxy
67
and
7 7 If
[r.] and tr._], i - 1,2, contain entries of [R ]
X IB gZ
In addition,
z 2
j]* i
=
12, contain the coefficents p. shown in Eq. (B.31)
z z
.,] and [ M ^O^
,
* * *^
c o n t a i n t h e
transverse leakage coefficients
introduced in Eq. (3.68),
and
[r_.], i
E
1,2, contain the axial boundary coefficients analogous to
those introduced in Eq. (2.89).
Applying the Gauss-Seidel procedure to Eq. (4.13) yields
{ r
xy, jout(m2) _
g
( n )
+ [r
xyz, jOu t ^- l )
+
r xyz, in,B(m2-l)
1 -gxy,odd ~ -gxy.odd 1 -gxy.evn IB
J
-gz
j m 2 )
m g
( n ) __ ,z,
L
(
+
[
r
] j
-gz,odd -gz,odd
p
l -g,odd 1 -gxy.odd
-
1
) + [r
2
] jin,B(m2-l)
gz.evn 1B
J
-gz
jOut(m2-l)
-gxy,evn
T
in,B(m2)
r
z ,
T
out,(m2) . , s ,
T
out(m2)
' L r., |l J ., T lt.l J
-gz Bl -gz,odd 2 -gz,evn '
m2 - 1,...,M2. (A.14)
68
Here m2 denotes the axial sweep index, and the outer iteration index n has
been dropped from the partial current vectors to simplify the notation. As
mentioned previously, two axial sweeps per group are performed at each outer
iteration, and thus
M2 E 2, g = 1.....G. (4.15)
o
It is clear that the matrix in Eq. (4.13) is reducible since the leakage
vectors L , , and L can be written in terms of the hex-plane partial
g O dd

g e v n
L

g,evn
current vectors J
O u
,, and J
u
. However, we prefer the form shown
-gxy.odd -gxy.evn
K
since, in the actual calculation, the leakage moments are not eliminated in
favor of the partial currents.
The matrices [r,
y
] and [r
2
y
] are block-diagonal with block sub-matrices
identical to the two-dimensional matrix [Rg] shown in Eq. (4.8). Therefore,
the equations represented by the first line in Eq. (4.14) can be decoupled
into separate equations for each of the odd-numbered planes. These single-
plane equations are identical in form to the two-dimensional global response
matrix equation, and are thus solved using the iteration shown in Eq. (4.9).
As in two dimensions, the number (Mlg) of iterations performed on the hex-
plane is calculated using Eq. (4.11).
In summary, the three-dimensional response matrix equations are solved
using M2g(=2) axial mesh sweeps in which first the odd-numbered planes and
then the even-numbered planes are processed during each sweep. The hex-plane
partial currents are computed using Mlg (typically 2) four-color checkerboard
sweeps on each plane, while the outgoing axial partial currents for the p.
,,
ane
are computed using a single (sequential) sweep of the nodes on the plane.
4.3 Coarse-Mesh Rebalance Acceleration of the Outer Iterations
The outer (fission source) iterations are accelerated using the well-
known coarse-mesh rebalance method ^ in combination with the asymptotic
source extrapolation technique discussed in Section 4.4. Coarse-mesh
rebalance has proven to be an effective means of accelerating the convergence
of iterative schemes encountered in the solution of the neutron transport,
neutron diffusion, and fluid dynamics problems. The basic idea of the method
is to scale the fluxes calculated at each outer iteration on the "fine-mesh"
by rebalance factors computed such that a neutron balance is enforced over
each cell (region) of a "coarse-mesh" super-imposed on the fine mesh. This
approach is nonlinear since the fine mesh fluxes are used to compute the
coefficients of the coarse-mesh equations.
69
4.3.1 Construction of the Coarse-Mesh Equations
The rebalance equations are constructed on a mesh defined such that
each ring of hexagons forms a coarse-mesh region in the hex-plane (see Fig.
4.3) and each fine-mesh plane is assigned to an axial coarse-mesh region
comprised of one or more adjacent fine-mesh planes. The coarse-mesh regions
are denoted by V
m
, m * 1,...,M, where
M = I J
and
I = number of rings of hexagons (including the central hexagon)
J = number of co. rse-mesh rebalance intervals in the axial direction.
Typically 2 or 3 axial planes are combined to form a single axial coarse-mesh
region. The use of rings of hexagons as hex-plane coarse-mesh regions simpli-
fies both the construction of the coarse-mesh equations and their solution
since, in two dimensions, these equations have the simple tri-diagonal
structure of conventional one-dimensional finite difference equations.
Fig. 4.3 Coarse-Mesh Rebalance Regions in the Hex-Plane
70
Th e rebalance equations are derived In th e following manner* Using
th e nodal flux es, interface partial currents, and eig envalue computed at
th e nth outer iteration, th e following coarsemesh balance equation is
obtained by multiplying th e th reedimensional nodal balance equation
[(Eq. (3.1)] by V^
>
and th en summing th e result over all nodes k con
tained in coarsemesh reg ion V and over all energ y g roups g :
J + [> J + A J T P , m = 1 M. (4.16)
Here th e integ rated effective absorption rate and th e integ rated
production rate for coarsemesh cell m are
A
m
"
E v
k
E
keV
m
and
(4.18)
keV
m
g = 1 g
wh ere th e notation keV"
1
implies all nodjs k such th e V e V
m
. Th e
effective absorption cross section is defined by*
.a,k _ r.r,k _
''g
=
g Lu " g 'g
g '
s k
wh ere J is th e scattering cross section from g roup g to g roup g '.
Th e quantity J
m
represents th e total neutron leakag e from reg ion m to
its neig h boring coarsemesh reg ion ; th e summations in Eq. (4.16) are
over all such adjacent coarsemesh reg ions. As sh own in Fig . 4.3, each
*Th e " effective" absorption cross section defined in Eq. (4.19) is not
necessarily equal to th e " true" absorption cross section because th e D IF3D
scattering cross sections include contributions due to (n,2 n) reactions.
71
coarse-mesh region (with the obvious exception of the central hexagon)
imbedded in the reactor has two neighbors in two dimensions and four
neighbors in three dimensions* The leakages are readily expressed in
terms of the available face-averaged partial currents across the surface
m&
S common to coarse-mesh regions m and i :
S
k
(r ) V j
o u t
'
k
<
n
>(
r
),
k e
v
m
. (4.20)
s L.J g s
The notation here differs slightly from that used previously. J.
u
' (r
s
)
is the face-averaged outgoing partial current from node k across a nodal
surface (denoted by local coordinate r
Q
) which forms part of the surface
S , and S
k
(r
s
) is the area of the nodal surface. The total leakage J
is written in a similar manner in terms of outgoing partial currents from
nodes in coarse-mesh region V*.
Equation (4.16) will not be satisfied if the outer iteration procedure
is not converged because the fluxes and partial currents were calculated
using a fission source from the previous iteration. We can, however,
improve the solution (and hence the overall convergence rate of the outer
iterations) by first defining the "rebalanced" solution
-k(n) _ m -k(n)
g g
rout,k(n), * _
f
m -jout,k(n), .
g s' g s'
1,...,G (4.21a)
k e V
m
, (4.21b)
and then calculating the rebalance factors f , m = 1,...,M such that
the rebalance solution satisfies Eq. (4.16). The following eigenvalue
equation for the rebalance factors is thus obtained:
f* + [ X ) ^
% +
A*! f
m
" J P
m
f
m
, m - 1.....K. (4.22)
Equation (4.22) is solved for the rebalance factors and the new estimate
for the eigenvalue as described in the following section. Note that the
rebalance factors will approach 1 as the outer iterations converge.
72
A.3.2 Solution of the Coarse-Mesh Equations
The coarse-mesh equations represented by Eq. (4.22) can be combined
in the form
[M]f = J [P]f , (4.23)
where
f 5 col [fJ...,f
M
]
[P] i diag [pJ...,P
M
] ,
and the rebalance equations are ordered first by axial region and then
by hex-plane region. Therefore, [M] has the following block tri-diagonal
structure:
[M] =
[A
2
] [B
2
] [C
2
]
[A
3
] [B
3
] [C
3
] (4.24)
The entries of the diagonal matrices [A-jJ and [C-j_] involve the coarse-
mesh leakages [Eq. (4.20)] in the hex-plane, while the off-diagonal
entries of the tri-diagonal matrices [BjJ involve leakages between
adjacent axial coarse-mesh regions. These sub-matrices are square with
dimension J, the number of coarjse-mesh regions in the axial direction.
The sub-matrices in Eq. (4.24) are reduced to scalar quantities in two-
dimensional calculations.
Equation (4.23) is constructed and solved following each outer
iteration. The solution to this eigenvalue problem can be obtained using
either the power method
21
or the Wielandt method
21
of fractional iteration.
For problems in which the [M] matrix can be inverted directly, the Wielandt
method is often more efficient for reasons which will be discussed below.
This approach is based on the application of the power method to the
"shifted" eigenvalue problem obtained by rewriting Eq. (4.23) in the form
73
lM]f - i [P]f , (4.25)
A
where
[M] = [M] - i - [P] (4.26)
A
e
A e
and A
e
is an estimate for the fundamental-mode eigenvalue A
o
(k-effective)
such that A
e
> A
o
. This estimate is calculated using
A
g
= 1.05 A^
11
"
1
\ (4.28)
where A is the rebalanced eigenvalue computed at the previous outer
iteration. The convergence rate of the power method is determined by the
dominance ratio* of the matrix [M
- 1
][P]; the closer this ratio is to 1,
the slower the convergence rate. It can be shown that for A
g
> A
o
, the
dominance ratio of [M ]"
1
^] is smaller than that of [M]
- 1
[P]. Hence the
Wielandt method, which is obtained by applying the power method to Eq.
(4.25), will converge faster than the power method applied directly to
Eq. (4.23). "e thus solve Eq. (4.25) using the following iterative
procedure:
S
( t )
E [M]
_1
[P] f
(t _1 )
(4.29a)
(4.29b)
(4.29c)
lit?] t^- 'X
The dominance ratio of a matrix is defined by
a =
where A
Q
and A. are, respectively, the largest and second largest
eigenvalues of the matrix.
74
where t is the coarse-mesh Iteration index and |r||| denotes the L.
vector norm. The indicated matrix inversion is performed using a
direct matrix factorization technique
21
based on the block tri-diagonal
structure of [M]. The factorization is performed prior to the start of
the coarse-mesh iterations; a forward elimination, backward substitution
procedure is then used to compute the rebalance factors at each coarse-
mesh iteration. A fixed (user-specified) number of coarse-mesh iterations
are performed at each outer iteration. Numerical calculations to date
have demonstrated that due to the efficiency of the Wielandt method, only
two coarse-mesh iterations are required at each outer iteration.
The computed rebalance factors are used to scale the partial
currents and fission source moments in accordance with Eqs. (4.21):
jout,k(n)
s
m out,k(n) (4.30a)
- 8 - 8
jk(n) _
f
m ^(n )^
R y
m
# (4 #3 0 b )
The fission source is scaled prior to checking the convergence of the
fission source (see Fig. 4.1), while the group partial currents are
scaled at the beginning of the loop over groups in the next outer
iteration in order to avoid an additional group loop following the
rebalance procedure. (It is not necessary to scale the flux moments
<|> because, as shown in Eq. (4.1), they do not enter into the
calculation of the group source term Q if there is no up-
scatter). The final estimate for the eigenvalue at the n-th outer
iteration is obtained from Eq. (4.27)
where A is the final iterate computed during the coarse-mesh iterations.
4.4 Acceleration of the Outer Iterations Using Asymptotic Extrapolation
The outer iterations are also accelerated using an asymptotic source
extrapolation procedure similar to that described in Ref. 23. This procedure
is based on the assumption that the fission source converges to the exact
(fully converged) solution ijX
00
) with the asymptotic behavior
+ R c
n
, (4.32)
75
where is the global fission source vector containing the node-averaged
fission sources for all nodes, n is the outer iteration index, a is the
dominance ratio (see footnote in Section 4.3.2) and R is an unknown
vector. The vector ijj(
n
) is constructed using the rebalanced fission
source [Eq. (4.30b)] computed at the end of the n-th outer Iteration. An
estimate for the dominance ratio can be calculated in a manner consistent
with Eq. (4 . 3 2):
.(n)
(4.33)
where II* || denotes the L vector norm. Given the assumed asymptotic
behavior shown in Eq. (4 . 3 2), an improved estimate $ *
s
(4.34)
where
(0
(n) _
(4.35)
The extrapolation shown in Eq. (4.34) is applied only when asymptotic
behavior is observed, as determined by the criterion
min < 0.1, (4,36)
where
(n) .
b>
(n)
(4.37)
An additional criterion which must also be satisfied prior to extra-
polation is
n - n * > 5, (4.38)
where n denotes the outer iteration at which the most recent
extrapolation was performed*
76
The convergence of the fission source is checked following the
rebalance step as shown in Fig. A.I. If Eqs. (4.36) and (4.38) are
satisfied, the fission source moments and interface partial currents
are extrapolated in accordance with Eq. (4.34):
*out,k(n) _
T
out,k(n) ,
-g -g -g "g
g
. j
G
k=1,...,K
# ( 4 # 3 9 )
As with the rebalance scaling of the partial currents shown in Eq.
(4.30b), these operations are performed during the following outer
iteration. The rebalance factors are applied prior to the extra-
polation, and thus the vectors appearing on the right hand side of
Eq. (4.39) are the most recent rebalanced solutions.
4. 5 On the Computational Efficiency of the Nodal Scheme
DIF3D finite difference calculations are typically performed using 6
triangular mesh cells per hexagonal fuel assembly and 3 axial mesh planes per
axial burnup region, for a total of 18 flux unknowns per group per hexagonal-z
cell defined by the axial burnup boundaries. As will be shown in the following
section, the accuracy of the nodal scheme permits the use of a single mesh
plane per axial burnup region. The nodal option thus involves a total of 13
principal unknowns per group for this same hexagonal-z cell. Numerical results
given in the following section demonstrate that in spite of the similar numbers
of unknowns involved in the two schemes, the nodal option runs approximately
8 times faster than the finite difference option. This rather surprising
improvement in efficiency can be attributed to differences in the two solution
algorithms. The nodal solution procedure described in Sections 4.1 and 4.2
requires the recalculation of a total of 32 partial currents and flux moments
per node per group per outer iteration. A typical finite difference calcu-
lation using an average of 12 inner iterations per group requires calculation
of a total of 12 x 18 = 216 fluxes per coarse-mesh cell (node) per outer
iteration. Thus, measured in terms of the more meaningful number of total
unknowns computed during an outer iteration, the nodal option offers a
potential increase in efficiency by a factor of 216/32 s 6.5. Although the
actual improvement obviously depends on additional factors such as the CPU
time necessary to compute an unknown and the total number of outer iterations,
this predicted ratio is consistent with that observed in the numerical
comparisons given in the following section.
77
5. SAMPLE NUMERICAL CALCULATIONS
Numerical results are presented in this section for two computational
benchmark problems representing a 300 MWe homogeneous-core LMFBR and a
i.000 MWe heterogeneous-core design. Both problems are analyzed in two and
three dimensions. Nodal and finite difference results are compared in terms
of the accuracy of the respective spatial approximations and the computational
effort required to achieve this accuracy. The accuracy is measured by the
errors with respect to a spatially-converged solution of the multigroup
neutron diffusion equation, while the computational effort is reflected by
the central processor unit (CPU) time. In particular, we are interested in
the error in the computed value of k-effective,
k - k
r e f
eff
as well as the errors in the flux and/or power density averaged over specified
regions of the reactor. The reference solution (e.g. k
e
|f) is calculated
using Richardson extrapolation of the finite difference results, and is thus
assumed to be the exact solution of the multigroup diffusion equation. All
calculations were performed using the IBM 370/195 computer at Argonne National
Laboratory (ANL) with the exception of the three-dimensional LCCEWG calcula-
tions, which were done on the ANL IBM 3033 computer. The CPU times on the
3033 are 35 to 50% larger than the 370/195 times for the same problem. All
CPU times quoted here are for the calculation of the coupling coefficients and
the outer iterations, and do not include the time required for input processing
and solution edits.
5.1 The SNR Benchmark Problem
The SNR benchmark problem
2tf

2 5
is a 4-group model of a 300 MWe
homogeneous-core LMFBR originally specified in both Cartesian and triangular
geometry. The modified problem solved here is obtained by altering the
outer boundary of the triangular-geometry model (while preserving the volume
of the core) to allow imposition of boundary conditions along surfaces of
hexagons. The model consists of a two-zone core surrounded by radial and
axial blankets without a reflector. The height of the active core is 95 cm,
and each axial blanket is 40 cm thick. A total of 11 rings of hexagons
(including the central hexagon) are included in the model, with a lattice
pitch of 11.2003 cm. Vacuum boundary conditions are imposed on the outer
surfaces of the blankets. The full-core model includes a total of 18 control
rods, with 6 of these rods parked at the core-upper axial blanket interface,
and the remaining 12 rods inserted to the core midplane. All calculations
were performed using sixth-core planar symmetry.
78
5. 1. 1 Two-Dimensional Results
The two-dimensional problem solved here corresponds to the ori gi nal
"rods in" configuration
21
**
25
representing a horizontal s l i ce taken through
the upper half of the core. The NH2, NH3, and NH4 approximation shown in
Table 5.1 were obtained using the nodal option in DIF3D with N - 2, 3, 4
respectively in Eq. ( 2. 33) . The fi ni t e difference cal cul ati ons used the
indicated number of triangular mesh c e l l s per hexagonal assembly. The nodal
resul t s demonstrate the improved accuracy of the NH4 scheme rel at i ve to the
NH2 and NH3 approximations. With the exception of the average flux in the
radial blanket, the NH4 resul ts are considerably more accurate than the
DIF3D(6A) resul t s. In particular, the NH4 eigenvalue and average flux in
the control rod are almost as accurate as the DIF3D(24A) resul t s. Thus the
accuracy of the NH4 approximation fal l s between that of the DIF3D(6A) and
DIF3D(24A) resul t s, with a reduction in CPU time by factors of 1.6 and 8.2
rel at i ve to the fi ni t e difference cal cul ati ons.
Table 5.1 Summary of Results for the Tvo-Diuenslonal SNR Benchmark Problem*
Method
DIF3D(NH2)
DIF3D(NH3)
DIF3D(NH4)
DIF3D(6A)
DIF3D(24A)
Reference
k-eff
1.12753
1.12583
1.12529
1.12728
1.12475
1.12375
\>
0.34
0.27
0.14
0.31
0.09

-0.47
-0.39
-0.22
-0.42
-0.12

e
oc< *>
0.76
0.66
0.42
0.72
0.32

1.89
2.11
1.14
0.83
0.27
~
-1.51
-0.96
-0.44
-1.84
-0.51

CPU
Time (sec)
0.8
0.8
0.9
1.4
7.4

*
e
IC'
e
0C*
e
RB*
and e
CR
a r e t h e e r r o r B i n t ne
e
rou
P~ and region-averaged
fluxes for the inner core, outer core, radial blanket, and control rod regions,
respectively.
a
The reference solution is obtained by Richardson extrapolation of DIF3D(24A) and
DIF3D(96A) solutions.
5. 1. 2 Three-Dimensional Results
Results for the three-dimensional SNR benchmark problem are summarized
in Table 5. 2. The nodal cal cul ati ons used the NH4 hex-plane approximation in
combination with a cubic axi al approximation [N
z
= 3 in Eq. ( 3. 29) ]. The
cal cul ati ons with 8 and 18 axi al mesh planes used 4 and 10 mesh planes,
respect i vel y, in the acti ve core, and 2 and 4 mesh planes, respecti vel y, in
each axi al blanket. Extrapolated resul ts assuming an i nfi ni t e number of axi al
mesh planes have been included in order to allow i sol at i on of the errors due
79
to the axial approximations in the nodal and finite difference schemes. For
example, these results show that the 0.16% eigenvalue error in the 8-plane
nodal calculation involves contributions of 0.13% and 0.03% due to the hex-
plane and axial approximations, respectively. (This hex-plane error is
consistent with 0.12% error observed in the two-dimensional problem.) Similar
analysis of the finite difference results shows that the axial contribution to
the total eigenvalue error in the 18-plane and 36-plane calculations is 0.30%
and 0.07 - 0.08%, respectively. Similar trends are observed in the flux
errors, although there is some fortuitous cancellation of hex-plane and axial
errors in the finite-difference results for the inner core and radial blanket.
We conclude that the axial accuracy of the nodal scheme with 8 axial planes is
superior to that of the finite difference approximation using 36 planes.
Furthermore, although the overall accuracy of the 8-plane nodal calculation
is superior to that of the 36-plane 6 triangles-per-hexagon finite difference
results, the nodal calculation required a fa" _^ VL 6 less computing time than
this finite difference calculation.
Table 5.2 Su a u r y of Reculta for the Thrat-DlMiulonal SNR Banchaark Problea*
Method
DIF3D(NH4)
DIF3D(NH4)
DIF3D(NH4)
DIF3D(6A)
OIF3D(6A)
DIF3D(6A)
DIF3D(24A)
DIF3D(24A)
DIF3O<24A)
Reference*
No. of
Axial Planes
8
18
18
36
r
18
36
-
k-eff
1.01150
1.01125
1.01120
1.01505
1.01280
1.01205
1.01342
1.01118
1.01043
1.00989

K
(D
0.16
0.13
0.13
0.52
0.29
0.22
0.35
0.13
0.05

1C<
-0.17
-0.18
-0.18
-0.18
-0.27
-0.29
-0.05
-0.04
-0.08

oc<
0.23
0.22
0.22
0.52
0.42
0.38
0.23
0.13
0.09

0.9S
0.96
0.96
0.22
0.47
0.56
-0.20
0.05
0.14

-0.30
-0.11
-0.07
-2.55
-0.60
-0.06
-2.61
-0.64
0.02

e
CR
C)
-0.60
-0.44
-0.39
-2.56
-1.72
-1.44
-1.48
-0.64
-0.36

CFU
Time (aln)
0.2
0.6
0.6
1.6
3.1
6.0

*
C
IC '
C
0 C *
C
R B *
C
AB * *
nd C
C R *
re che e r r o r
*
l n t h e
group- and region-averaged fluxei for the Inner
core, outer core, radial blanket, axial blanket, and control rod regions, respectively.
*The reference aolutlon la obtained by Rlchardion extrapolation of the DIF3D(6A) - 18 plane and
DIf3D(24A) - 36 plane aolutlona.
5.2 The LCCEWG Benchmark Problem
The Large Core Code Evaluation Working Group (LCCEWG) benchmark
problem^
6
is a model of a 1000 MWe heterogeneous-core design with a lattice
pitch of 16.33 cm. The core layout is shown in Figs. 5.1 and 5.2. Additional
specifications for this problem are given in Ref. 27. Results summarized here
include both beginning-of-life (BOL) and depletion calculations [using REBUS-3
(Ref. 5)] for two- and three-dimensional models with four energy groups.
Detailed comparisons of 8-group BOL and depletion calculations using finite
difference, nodal, and flux synthesis'* neutronics solutions have been reported
elsewhere.
7
80
RADIAL
SHIELD
RAD IAL
BLANKET
DRIVER
INNER
BLANKET H I
DRIVER 31
INNER
BLANKET IE
DRIVER I
INNER
BLANKET I
BATCH LOADING S E QU E NC E - ALL ASSEMBLIES
LABELLED V ARE REPLACED AT THE END OF
IRRAD IATION CYCLE "N"
PRIMARY CONTROL ROD USED FOR CRITICALITY
AD JUSTMENTS DURING BURNUP
PRIMARY CONTROL ROD
S J SECONDARY CONTROL ROD
Fig. 5. 1 Sixty-Degree Sector of the Core Layout for the LCCEWG Benchmark Problem
AXI AL D I ME NS I O N, cm
H
NJ

1
o>
o
1
P>
cn
ro
1
09 ^
00
00
b>
ro
I
CD
H -
O
3
CO
l - h
O
ft
n>
s
8
I
-
O
W
E
R
A
X
I
i
\
L

S
H
I
E
L
o
s>
I
R
I
F
I
C
E
A
X
I
A
L
B
L
K
T
S
A
X
I
A
L
B
L
K
T
S
A
X
I
A
L
B
L
K
T
S
INNER BLANKET ZONE
D RIVER ZONE I
INNER BLANKET ZONE
DRIVER ZONE H
INNER BLANKET ZONE
DRIVER ZONE I H
RAD IAL BLANKETS
RAD IAL SHI ELD
I
n
IE
B
L
K
T
S
j
j
B
L
K
T
S

|
|

B
L
K
T
S

g
1

i
A
X
I
A
L

|
|

A
X
I
A
L

|
|

A
X
I
A
L

j
r>
r>
- n
I
S
S
I
O
N

G
A
S
P
L
E
N
U
8
so
o
o
oo
rn
33

o
30
j>
82
5.2.1 Two-Dimensional Results
Table 5.3 summarizes BOL results for the two-dimensional version of the
LCCEWG benchmark problem. The regions used in computing the errors are
defined by the zones shown in Fig. 5.1; thus there are three driver-fuel
regions, three inner-blanket regions, and one radial blanket region. The
average error in the driver fuel, for example, is calculated by averaging the
errors over the three fuel regions. The nodal results demonstrate that the
NH4 approximation is necessary in order to achieve acceptable accuracy in the
blankets. Although the NH4 error in the driver fuel is slightly higher than
the DIF3D(6A) error, the NH4 results for k-effective and the power densities
in the blankets are significantly better than the DIF3D(6A) results. This
improved accuracy is obtained with a reduction in computing time by a factor
of 2.5 relative to the DIF3D(6A) approximation.
Table 5.3 Sumary of BOL Results for the Two-Diaensional LCCEWG Benchmark Problem*
Method
DIF3D(NH2)
DIF3D(NH3)
DIF3D(NH4)
DIF3D(6A)
DIF3D(24A)
Reference
k-eff
1.01064
1.00957
1.01002
1.01323
1.01048
1.C0980

k< *>
0.08
-0.02
0.02
0.34
0.07

*DF<*>
1.54
1.55
0.82
0.69
0.15

e
IB
(Z )
7.53
6.37
2.96
5.50
1.23

3.13
2.33
0.98
3.31
0.63

CPU
Time (sec)
1.9
2.0
2.0
4.9
22.3

*- _ -
E
DF*
e
IB*
a nd e
RB
a r e t h e a v e r a
8
e
errors in the region-averaged power densities
over a ll regions contained in the driver fuel, internal blankets, and radial
blanket, respectively.
The reference solution is obtained by Richardson extrapolation of DIF3D(24A)
and DIF3D(96A) solutions.
Depletion results
2
*
1
for the two-dimensional model are summarized in
Table 5.4. The reactivity swings due to burnup over two non-equilibrium
cycles are shown as well as average burnups and breeding ratios computed at
the end of cycle 1 (E0C1). The average inner-blanket burnups computed using
the NH4 approximation are more accurate than the DIF3D(6A) results, while the
nodal burnup swings are as accurate as those computed using the DIF3D(24A)
option. It is clear that the improved inner-blanket burnups are a direct
consequence of the reduced inner-blanket errors shown in Table 5.3. The
improved inner-blanket solutions, in combination with the accuracy of the
eigenvalues computed using the nodal option, lead to very accurate predictions
of the reactivity swing due to burnup.
83
Table 5*4 Susaary of Depletion Results for the Two-Dleensionul
LCCEWG Benchmark Problem
k
eff
(BOCl) - k
eff
(EOCl)
k
e ff
(B0C2) - k
eff
(EOC2)
EOC1
Average Burnup (MWd/MT)
Inner Core
Middle Core
Outer Core
Inner Blanket
Middle Blanket
Outer Blanket
Breeding Ratio
i 2 Three-Dimensional
DIF3D(NH4)
0.0203
0.0169
2.99+4
3.11+4
2.68+4
2.05+3
4.94+3
3.12+3
1.326
Results
DIF3D(6A)
0.0219
0.0179
2.99+4
3.12+4
2.67+4
2.00+3
4.79+3
3.05+3
1.319
DIF3D(24&)
0.0207
0.0169
3.01+4
3.12+4
2.67+4
2.08+3
4.98+3
3.15+3
1.323
DIF3D(96A)
0.0205
0.0166
3.01+4
3.13+4
2.66+4
2.10+3
5.04+3
3.17+3
1.324
Table 5.5 summarizes results for the three-dimensional LCCEWG problem.
The average errors are calculated by averaging the errors for regions defined
in the hex-plane by the zones shown in Fig. 5.1 and in the axial direction by
the axial burnup regions. A total of 12 axial burnup regions are defined by 8
uniform (15.24 cm) axial mesh intervals in the active core and 2 uniform
(17.78 cm) intervals in each axial blanket. The 14-plane nodal calculation
used a single mesh plane per axial burnup region plus one additional mns'h
plane each in the lower axial shield and the fission gas plenum. The 28-plane
and 56-plane meshes are defined by uniform refinement of the 14-plane mesh.
As for the three-dimensional SNR benchmark problem, results assuming an
infinite number of axial planes are given in Table 5.5* Using these extra-
polated results, it can be seen that the errors in eigenvalue due to the axial
approximation are greater in the 56-plane finite difference calculations than
in the 14-plane nodal calculation. Comparison of the 14-plane nodal and 56-
plane D1F3D(6A) calculations shows that the nodal option produces somewhat
larger (but very acceptable) errors in the driver fuel, but yields signifi-
cantly smaller errors in eigenvalue and in the inner and radial blankets* The
larger errors in the fuel may be due to the flat approximation [Eq. (3.61)] to
the axial leakage. The increasing errors with decreasing axial mesh spacing
observed in the nodal results for the driver fuel and inner blankets are not
understood, although it is likely that this behavior is due to a fortuitous
cancellation of hex-plane and axial errors and/or uncertainties in the
reference solution. The nodal eigenvalue error is considerably larger than
for the two-dimensional version of this problem. Again, this is in part due
to the flat leakage approximation, but may also reflect the larger flux
gradients in the three-dimensional model due to partially-inserted control
rods* (The control rods are withdrawn in the two-dimensional model). Although
84
the overall accuracy of the 14-plane nodal calculation Is superior to that of
the 56-plane DIF3D(6A) calculation, the nodal calculation required a factor of
9 less computing time.
Table 5. 5 Summary of BOL Results for the Three-Dimensional LCCEWG Benchmark Problem*
Method
OIF3D(NH4)
DIF3D(NH4)
DIF3D(NH4)
DIF3D(6A)
DIF3D(6A)
DIF3D(6A)
DIF3D(24A)
DIF3D(24A)
DIF3D(24&)
Reference
No. of
Axial Planes
14
28
m
28
56
m
28
56
m
m
k-eff
0.99846
0.99829
0.99820
1.00329
1.00206
1.00165
0.99978
0.99855
0.99814
0.99697
c
k
<%)
0. 15
0. 13
0. 12
0.63
0.51
0.47
0.28
0.16
0.11

c
DF
(I)
1.4
i . 6
1.6
0.8
0.7
0.7
0.5
0.2
0.1

;
I B
<
3.1
3.2
3.2
4.2
4.4
4.4
1.1
1.1
1.1

0.8
0.7
0.7
3.4
3.1
3.0
1.3
0.8
0.6

2.3
1.6
1.5
5.3
2.1
1.0
4.9
1.3
0.1
.._
CPU
Tiee (aln)
1.1
2.5
5.3
10.2
30.5
59.0

*-
e e _, and e,_ are the average errors In the region-averaged power densities over all
DF
,
IB ..
regions contained in the driver fuel, inner blankets, radial blanket, and axial blankets,
respectively.
a
The reference solution i s obtained by Richardson extrapolation of the DIF3D(6A) - 28 plane and
DIF3D(24o) - 56 plane solutions.
Table 5.6 summarizes computed burnup swings over the f ir st half of a
388.5 day cy^le. The burnup swing computed using the nodal option is very
accurate as shown by the excellent agreement with the DIF3D(24A) r esult .
Thus, as in two dimensions, the improved accuracy of the eigenvalues and
blanket fluxes computed with the nodal option leads to significantly
improved predictions of the react ivit y swing due to burnup.
Table 5.6 Computed Burnup Swings for the Three-Dimensional
LCCEWG Benchmark Problem*
Method
Number of
Axial Planes
Burnup
Swing (Ak)
Neutronics
CPU Time (min)
DIF3D(N H 4)
DIF3D(6A)
DIF3D(24A)
17
42
42
-0.00426
-0.00489
-0.00420
2.6
14.0
31.1
*0ver the first half of a 388.5 day cycle.
85
From a practical viewpoint, the total dollar cost is an important
measure of the computational efficiency of any numerical scheme proposed for
the solution of large-scale reactor physics problems. The job cost typically
includes contributions due to core storage and I/O activity in addition to the
CPU time. Of course, the* weights assigned to these contributions vary from one
installation to another. Table 5.7 shows the core storage requirements, EXCP
total (which is a measure of I/O activity), and dollar cost at ANL of several
calculations for the three dimensional LCCEWG problem. The nodal storage
requirements shown in Table 5.7 include the storage necessary for all coupling
coefficients required during the depletion calculation. The reduced storage
requirements for the nodal option are due primarily to the need to store
coupling coefficients only for unique nodes as described in Section 3.5. The
nodal calculation was run with all data for one energy group in core, while
the finite difference calculations core-contained only a fraction of this data
in an effort to reduce the total job cost. Thus the nodal EXCP count is
significantly smaller. This reduced I/O activity, in combination with the
smaller core storage requirements for the nodal option, leads to even greater
reductions in dollar cost than those already observed in CPU time. For
example, relative to the DIF3D(6A) calculation with 56 planes, the nodal
calculation required a factor of 9 less CPU time with a reduction in dollar
cost by a factor of nearly 12.
Table 5.7 Summary of Execution Statistics for the Three-Dimenpional LCCEWG Problem*
Method
DIF3D(NH4)
DIF3D{ 6A)
DIF3D(6A)
DIF3D(24A)
No. of
Axial Planes
14
28
56
28
L
l
116
297
589
1188
L
2
215
852
1693
3412
L
3
116
223
215
340
CPU
Time (min)
1.1
5.3
10.2
30.5
EXCP Total
( x l O
- 3
)
3.6
30.1
65.4
292.5
Cost($)
7.28
43.14
85.27
336.72
BOL configuration, 4 energy groups. The core storage does not include the storage
required by the DIF3D load module.
l^ = core storage [K-(REAL*8) words] required to contain all data for one group.
L
2
= core storage [K-(REAL*8) words] required to contain all data for all groups.
L~ = core storage [K-(REAL*8) words] actually used.
5.3 Calculations of Peak Power Densities
The accurate calculation of peak power densities using a nodal
formulation is limited by the lack of information concerning the spatial
distribution of the flux within the node. A simple procedure is used in the
nodal option to compute more accurate peak power densities than those obtained
by sampling only the available node-averaged values. The peak values are com-
puted in two-dimensional problems by sampling both the surface-averaged and
node-averaged values of the power density* This scheme is extended to three
dimensions by assuming the flux is separable in the hex-plane and axial
directions* Additional details of this procedure are provided in Appendix D.
86
Peak power densi t i es calculated for the three-dimensional SNR benchmark
problem are shown in Table 5*8. The fi ni te-di fference peak values are computed
by sampling both the surface and cel l -centered values of the power density.
The surface values are obtained in a manner consistent with the mesh-centered
fi ni t e difference formulation in D1F3D. The nodal resul ts calculated using 8
axi al planes agree with the reference sol uti on to within 1.0%. Although the
procedure used to compute peak values in the nodal option i s particularly
simple, the resul t s in Table 5.8 demonstrate that the accuracy of t hi s scheme
i s comparable to that achieved in fine-mesh fi ni t e difference cal cul ati ons.
Table 5.8 Computed Peak Power Densities for the Three-Dimensional SNR Benchmark Problem*
Method
DIF3D(NHA)
DIF3D(NH4)
DIF3D(6A)
DIF3D(6A)
DIF3D(24A)
D1F3D(24A)
Reference
8
No. of
Axial Planes
8
18
18
36
18
36
m
Inner Core
3.031-6
3.030-6
3.010-6
3.019-6
3.030-6
3.040-6
3.05 -6
Peak Power Densities (watts/cm
3
)
Outer Core
2.896-6
2.907-6
2.890-6
2.900-6
2.892-6
2.903-6
2.90 -6
Radial Blanket
2.683-7
2.678-7
2.633-7
2.650-7
2.635-7
2.652-7
2.66 -7
Axial Blanket
1.754-7
1.758-7
1.789-7
1.763-7
1.804-7
1.777-7
1.77 -7
*The power densities are normalized to a total power of 1 watt over the third-core model.
a
The reference solution is obtained by approximate extrapolation of the finite difference
results.
5.4 Overview of the Numerical Results
The results presented in this section have shown the accuracy of the
nodal scheme to be superior to that of the standard DIF3D finite difference
option using six triangular mesh cells per hexagonal fuel assembly. For
three-dimensonal calculations, the higher-order axial approximation in the
nodal scheme permits the use of an axial mesh which is at least 4 times
coarser than that used in a typical finite difference calculation. Particular
improvement is seen in the average fluxes in the blanket region (where the
largest finite difference errors typically occur) and in the computed values
for k-effective. This enhanced accuracy leads in turn to more accurate
predictions of inner blanket burnups, breeding ratios, and burnup reactivity
swings.
Relative to the standard finite difference option, the improved accuracy
of the nodal option is obtained in CPU times which are roughly 2 times smaller
in two-dimensional applications and 8-10 times smaller in three-dimensional
calculations. The reduced storage requirements for the nodal option can lead
to even greater reductions in the total cost of a calculation. Thus, in
87
summary, the superior axial accuracy of the nodal scheme permits the use of
a single axial mesh point per axial burnup region with a potential order-of-
magnitude cost reduction relative to the standard DIF3D neutronics option.
Although less dramatic cost reductions are obtained relative to the flux
synthesis module SYN3D (Ref. 4 ) , the nodal option most likely will supercede
the flux synthesis option due to its increased accuracy and ease of use*
88
6. USER INFORMATION
The nodal option in DIF3D is executed in essentially the same manner as
the standard finite difference option. This section highlights information
of particular interest to users of the nodal option, and is thus intended to
supplement the detailed documentation provided in Section 3 of Ref. 3 and in
the descriptions of the BCD input files.
6.1 Code-Dependent Input - A.DIF3D
Calculational parameters, storage containers, and edit sentinels are
specified via the BCD input file A.DIF3D. A description of thic file is given
in Appendix E. The edit sentinels on card type 04 and the convergence criteria
on card type 05 are directly applicable to the nodal option.* The only entry
on card type 06 which is used in the nodal option is the steady-state reactor
power in columns 49-60. A.DIF3D card types 07, 08, and 09 are not relevant to
the nodal option.
6.1.1 Data Management Options and Container Sizes - A.DIF3D Card Type 02
2 9
DIF3D uses the BPOINTER package to manage the dynamic allocation of
all variably-dimensioned arrays used in the code. These arrays are actually
stored in two large blocks of workspace called container arrays. The lengths
of the container arrays are specified on card type 02 of A.DIF3D. The FCM
(fast core memory) container is stored in fast core memory on both one-level
(e.g. IBM 370) and two-level (CDC 7600) computers. The ECM (extended core
memory) container is in fast core memory on one-level computers and in large
core memory (LCM) on two-level computers.
The specification of the FCM container length for the nodal option is
straightforward. On one-level machines, a relatively small allocation (e.g.
10,000 words) is sufficient since only small arrays are stored in this con-
tainer. On two-level computers such as the CDC 7600 computer, the length of
the FCM container should be specified as the maximum number of words available
in the small core memory. The number of FCM words actually used is given in
the data management edit which appears immediately before the outer iteration
history in the DIF3D output. A sample data management edit is shown in Fig.
6.1.
The length of the ECM container array determines the locations of the
files shown in Fig. 6.1. The minimum ECM container length is given by the
storage necessary to core-contain the group-independent files (e.g. fission
source moments, composition map) plus 1-group ECM buffers for the group-
dependent files stored on disk. Any additional available storage is used to
*Peak power densities and peak total fluxes included in the DIF3D edits are
computed using the procedure described in Appendix D,, The peak values are
written on an interface file PKEDIT which, in the ANL modular version of
DIF3D, is processed by the SUMMARY module.
DXF9B 4.0 3/82 CRBR HOTEL' 120 0E6REE SYMMETRY 13 AXIAL PLANES 20 GROUPS 9/03/82 1216.400 PA6E 71
H NM DIF3D (NODAL OPTION) STORAGE ALLOCATION M M
NUHBER OF HOROS IN DATA STORAGE CONTAINER =
MINIMUM NUMBER OF HOROS REQUIRED TO RUN THIS PROBLEM
HITH ALL OATA FOR 1 GROUP IN CORE
HITH SCATTERING BAND OF FLUXES IN CORE *
HITH ALL FILES IN CORE (OURING OUTER ITERATIONS) =
HITH ALL FILES IN CORE (DURING EDIT OVERLAY) *
FCH
6000
5991
5991
5991
5991
ECH
177000
112319
17600ft
943664
1032375
LOCATION OF SCRATCH FILES DURING OUTER ITERATIONS
RECORD
LENGTH
10205
12896
4396
12896
4396
361
624
RECORD
LENGTH
2197
18369
10205
12896
4396
FILE CONTENTS
FLUX MOMENTS
HEN HEX-PLAHE PARTIAL CURRENTS
MEU AXIAL PARTIAL CURRENTS
OLD HEX-PLANE PARTIAL CURRENTS
OLO AXIAL PARTIAL CURRENTS
CROSS SECTIONS
NODAL COUPLING COEFFICIENTS
LOCATION OF
FILE CONTENTS
NODE-AVERAGE FLUXES
FLUX SHAPE COEFFICIENTS
FLUX MOMENTS
NEH HEX-PLANE PARTIAL CURRENTS
NEH AXIAL PARTIAL CURRENTS
N O . OF
RECORDS
20
20
20
20
20
20
20
SCRATCH FII
NO. OF
RECORDS
20
20
*-* 20
20
20
FILE
LENGTH
204100
257920
87929
257920
87920
7220
12480
FILE
LENGTH
43940
367380
204100
257920
87920
LOCATION
DISK
DISK
DISK
DISK
DISK
DISK
DISK
LOCATION
CORE
DISK
DISK
DISK
OISK
RECORDS
IN CORE
11
1
1
1
1
1
1
RECORDS
IN CORE
20
1
1
1
1
00
TOTAL NUMBER OF HOROS USED FOR THIS PRCBLEH 5991 176008
Fig. 6.1 Sample Data Management Edit for the Nodal Option
90
core-contain the scattering band of fluxes followed by as many of the group-
dependent files shown in Fig. 6.1 as possible. Separate data management
strategies are used during the outer iterations and in the edit overlay.
Details of these strategies are given in Section 7.2.
The optimal choice of data management strategy (and thus the specification
of the ECM container length) depends upon the relative weights assigned to core
storage and 1/0 activity in computing job costs at a particular installation.
Experience at ANL suggests that two-dimensional problems should be run using
the container array sizes for all data in core during the edit overlay. The
job cost for three-dimensional calculations at ANL is generally minimized when
the scattering band of fluxes is core-contained. Note that the sample three-
dimensional problem shown in Fig. 6.1 was run using this data management mode.
Execution is terminated immediately following the edit of the data
management information if the minimum required storage exceeds the container
lengths specified on A.DIF3D card type 02. The user can thus specify
relatively small container arrays for the purpose of obtaining the data
management edit, and then use the information in this edit to specify the
appropriate container sizes for a subsequent run.
6.1.2 Nodal Option Parameters - A.DIF3D Card Type 10
The parameters for the nodal option are specified en card type 10 as
shown in Appendix E. It is strongly recommended that the default values for
these parameters be used.
6.1.3 Axial Coarse-Mesh Rebalance Boundaries - A.DIF3D Card Type 11
The coarse-mesh rebalance acceleration described in Section 4.3
requires specification of the axial boundaries which define the rebalance
mesh. The choice of these boundaries represents a trade-off between the
decreased number of outer iterations obtained with a relatively large number
of axial rebalance regions and the increased computational overhead required
for the solution of the rebalance equations. It is recommended that an axial
rebalance mesh no greater than 35 cm be specified in the active core and axial
blankets. In general, it is better to use a relatively fine axial rebalance
mesh since any reduction in the number of outer iterations typically offsets
the additional CPU time required for the solution of the increased number of
rebalance equations. It is also important that the axial rebalance mesh be
as uniform as possible since specification of adjacent rebalance regions with
very different axial dimensions can cause the outer iterations to diverge.
6.2 Geometry Input - A.NTP3
The geometry of the computational model is described via the BCD input
file A.NIP3. The card types discussed in the following sub-sections may
require attention when using the nodal option. Descriptions of these card
types are given in Appendix F.
91
6.2.1 Problem Geometry Specification - A.NIP3 Card Type 03
The nodal option is invoked by the following geometry-type sentinels:
110
114
116
120
124
126
Hexagonal,
Hexagonal,
Hexagonal,
Hexagonal-z,
Hexagonal-z,
Hexagonal-z,
Full
Sixth
Third
Full
Sixth
Third
Core
Core
Core
Core
Core
Core
Symmetry
Symmetry
in Plane
Symmetry
Symmetry
in
in
Plane
Plane
6.2.2 External Boundary Conditions - A.NIP3 Card Type 04
The nodal option does not treat reflectional symmetry in the plane; all
fractional-core models must be specified using rotational (periodic) symmetry
boundary conditions in the hex-plane* Thus, when geometry types 114, 116,
124, or 126 are specified on card type 03, boundary condition type 7 must be
specified in columns 13-18 of card type 04.
A single boundary condition type is imposed on all surfaces forming part
of the outer reactor boundary in the hex-plane. This boundary-condition type
is selected from appropriate external boundary conditions (i.e. either zero
flux or extrapolated boundary conditions) specified in the x- and y-directions.
Thus, consistent with obvious physical considerations, only one boundary con-
dition type should be specified on the outer reactor boundary in the hex-plane.
6.2.3 Variable-Mesh Structure - A.NIP3 Card Type 09
As in the triangular-geometry models, the axial mesh spacings for
hexagonal-z models are specified on card type 09. Results given in Section 5
have shown that the higher-order axial approximation in the nodal scheme
permits the use of an axial mesh which is at least 4 times coarser than the
5 cm axial mesh typically used in finite difference calculations. For most
burnup (REBUS-3) calculations, the axial mesh structure should coincide with
the axial region assignments on the A.NIP3 type 30 cards.
6.2.4 Location of Regions - A.NIP3 Card Type 30
Hexagons are assigned to regions via the A.NIP3 type 30 cards des-
cribed in Appendix F. Note that the hexagonal-geometry solution domains for
sixth- and third-core symmetries are rotated 30 degrees from the respective
domains used in the DIF3D triangular-geometry models* A special procedure
has been implemented in the input processor GNIP4C (Ref. 3) to facilitate
conversion of existing type 30 cards for most (but not all) finite difference
models to those required for the nodal option. When hexagonal geometry and
periodic boundary conditions are specified, GNIP4C will use the periodicity
(rotational symmetry) to assign hexagons not referenced on type 30 cards (but
included in the hexagonal-geometry solution domain) to appropriate regions*
The procedure requires that the hexagons bisected by the. triangular-geometry
92
fractional-core symmetry lines be assigned to regions in a manner consistent
with rotationally symmetric boundary conditions* (Recall from the discussion
in Section 6.2.2 that the nodal option treats only rotational symmetry in the
p l a n e ) . Thus, if the original triangular-geometry model was constructed using
boundary conditions type 3 (reflectional symmetry) on A.NIP3 card type 04, the
user must make certain that the region assignments are also consistent with
the rotational symmetry assumed in mapping the triangular-geometry model to
the hexagonal-geometry model. As an example, consider the following free-
format A.NIP3 cards included in a finite difference deck:
03
04
30
30
equii
03
04
30
30
70
3 2
REG1
REG2
red spe
114
7 2
REG1
REG1
3 2
3 2
3 12
cirlc
0 2
3 2
3 12
Note that the two region assignments (for ring 3, cells 2 and 12) must agree
in the nodal deck since rotational symmetry is specified on card type 04. If
this change in region assignments were not made, REG2 would be assigned a zero
region volume since it lies outside the hexagonal-geometry region of solution
shown in Appendix F. This would in turn give a fatal error in the REBUS-3
code. In summary, while the GNIP4C conversion procedure is very convenient
for converting large, previously-created finite difference decks to the nodal
option, the user must exercise some caution whenever it is invoked. For this
reason, warning messages are printed indicating new region assignments
generated by this procedure, as well as the names of any regions with zero
volumes. Printer-plotter and graphics maps for the nodal option are edited
using the triangular-geometry orientation; they may, therefore, be of marginal
use in checking input.
6.2.5 Background Region Name - A.NIP3 Card Type 31
The user is encouraged not to specify a background region since this
results in an unnecessary increase in the region of solution. If no type 31
card is present, external boundary conditions will be imposed along the outer
reactor boundary as mentioned under card type 04.
6.3 Restart Procedure
Analogous to the procedure in the standard finite difference option,
nodal calculations can be restarted by saving the NHFLUX interface file
(logical unit number 23 ) and placing it in BLOCK-OLD for a subsequent run.
93
6.A Limitations of the Nodal Option
The nodal option has the following limitations:
(1) Reflective boundary conditions are not permitted in the plane.
(2) Fixed source problems are not permitted.
(3) Adjoint calculations are not permitted.
(4) Internal black boundary conditions are not permitted.
(5) Up-scattering is not permitted.
94
7. PROGRAMMING INFORMATION
The information provided in this section is intended to assist users who
wish to make changes in the nodal option or who wish to understand the flow
of the calculation. Users faced with the task of making the DIF3D code
operational at their installations should refer to Section 5 of Ref. 3.
7.1 Programming Structure
Figure 7.1 shows the structure of the DIF3D code block.* The subroutines
preceded by an asterisk and the common blocks beginning with "NH" ("nodal hex")
are used only when the nodal option is invoked. All common blocks appearing
in the DIF3D code block are documented in this dummy subroutine BLOCKS included
in the DIF3D source code.
Since the functions of the subroutines shown in Fig. 7.i are described in
the comment cards which appear at the beginning of each subroutine in the code,
only a brief overview is provided here. The subroutine NHINIT called from
SSINIT controls the preliminary processing required before the data management
strategy can be determined in subroutines NHCORE and NHDISK. The preliminary
calculations include redefining the spatial mesh in the ordering used in the
nodal option, setting up pointers to partial currents, and determining the
number of unique nodes for which coupling coefficients must be computed and
stored. These tasks are discussed below in Sections 7.4.1, 7.4.2, and 7.4.3,
respectively.
The nodal flux calculation (including the calculation of the coupling
coefficients) is controlled by the primary overlay NHSST. Four secondary
overlays are invoked from NHSST. The secondary overlay DNHCCC controls the
calculation of the nodal coupling coefficients in two and three dimensions via
calls to NHCC2D and NHCC3D. The transverse leakage coupling coefficients are
calculated in NHTVLC, and the number of inner iterations (4-color checkerboard
sweeps) is determined via calls to NHINNR.
The secondary overlay DNHSTT initializes the flux and partial current
vectors by either reading a previously written NHFLUX interface file, or by
assuming a spatially flat flux shape in each energy group. These initiali-
zations are performed by FXREAD and FXINIT, respectively. The initial fission
source vector is then calculated via calls to FSINIT.
The secondary overlay DNHOUT controls the outer iterations in the nodal
option. The roles of the five principal subroutines (OUTR1,...,OUTR5) called
by this subroutine are shown in the flow chart in Fig. 4.1.
The final secondary overlay invoked by NHSST is DNHFIN, which performs
some final tasks required before the DIF3D edit overlay (DSSTOU) is invoked.
These operations include reordering the nodal fluxes in the GEODST ordering
*As discussed in Ref. 3, the DIF3D system consists of a collection of large
independent code blocks logically connected by a small "driver" subroutine.
The code block DIF3D performs the neutron flux and criticality calculations.
95
DIF3D
LINKR1
LINKR2
AREAS
VOLUME
GETBND
REVRSE
WDIF3D
UTILITIES
....
DEFICF
PURGCF
OPENCF
CLOSCF
STATCF
FEQUAT
FLTSET
IN2LIT
LINK
ERROR
SEEK
CRED
PCRED
BLKGET
BLKPUT |
OPEKDF |
CLOSDF |
DOPC |
IEQUAT |
INTSET |
LINES |
TIMER |
OPENDS |
REED |
DRED |
PNTGET |
BPOINTER |
COMMON
AM AY
.JNTRL
IOCOM
VBRNUM
DEBUG
BALBUF
PTITLE
JBCNTL
MHIOPC
BLOCKS |
SPECS |
IOPUT |
IOCOMC |
IOCOMD |
EDITDM |
PTITLE |
NHSFCM |
NHIOCM |
MHIOPD |
1START
I--BININP
I
I--SSINIT
I EDITCR
I
-SSTATE 1
OSWEEP |
PSWEEP |
TSWEEP |
ROWSRC |
TRISRC |
SORINV |
SCTSRC |
TOTSRC I
FISSRC |
Z EROBA I
OUTEDO |
OUTEDT I
CHEBE |
DACOSH
FILCPY
IK9IF3D
|~1MBEL
IRSEARC
i R P U P W * .
1
1 R G EOD S
|~RRTFLX
|-*RNHFLX
IRFIXSR
I F D INIT
1 Z M INIT
- *NHINIT
^^^^^AU OU in
1 F OR M SH
|-*FORMCM
ISSCORE
1 SSDISK
IINEDIT
|FORMMZ
|-*NHGEOM
|-*HEXMAP
I-*NHZ MAP
i-*NHPNT
|-*NHCCPT
| -*NHINED
|-*NHCORE
|-*NHDISK
XSINIT 1 --XSGET1
D X SR EV
- - D F D C AL
1XSGET2
XSREV
I OR T F D C
ITRIFDC
i , n p p F c i _
--DORPES 1 OR P ES2
n n i T T D i -
IRFLXIN
IFSRCIN
n i i T P P i ii mi
-ii rtlfTPIJ^
REGHAP
*GETIJ
XSEDIT
.flDPTMl
U KIr XHl
ORPIN2
IINNER1
- 1 C HEB Y1
ISCTSD2
|~T OT SD 2
- I INNER 2
IFISSD2
ICHEBY2
Wig, 7.1 Subroutine Map for tht DIF3D Cod* Block
96
1
1
1
1
1
1
1
1
1
1
|-*NHSST
| *NHOED0
j *INVERT
I *NHXSEC
1
I
I
1
1
1
I
I
I
IDSSTOU
TWODPR
TWODTB
BRED
DSEQUA
SCALPK
WPKEDT
*NHSHAP
*NHPEAK
*NHPKED
|-*DNHCCC
-*DNHSTT
1
|
1
I
I
-*DNHOUT

-*DNHFIN
DSSTO1
DSSTO2
DSSTO3
ADDVEC |
DIWEC |
-*NHCC2D
-*NHCC3D
-*NHTVLC
-*NHINNR
-*FXREAD
-*FXINIT
-*FSINIT
-*OUTR1
-*OUTR2
-*OUTR3
-*OUTRA
-*OUTR5
-*NHEDDM
-*NHVOL
-*FXSHAP
--FORMMR
B K L W G T
S S T O U 1
--POMINT
--EDCORE
S S T O U 2
--BALINT
F L X I O T
F L X R Z
|-*ACCEL
-*ACCL3D
-*LEAK3D
|-*SRCFIS
|-*SRCSCT
|-*SRCREX
| -*PCHEX *PCHEXB
|-*SRCZ1
-*SRCZ2
1 -*pcz *PCZ B
|-*FLXHEX
|-*FLXZ
|-*FSUPDT
I -*CMTRX 1 -*BKRING
|-*AXLEAK
|-*CMSOLV
|-*FSERRN
|-*CONVCK
*CPYFIL
O R T S R F
--TRISRF
W P O W E R
R P W A D D
--APWADD
--WFLUX
-*WNHFLX
R P S A D D
O R T B A L
T R I B A L
-*HEXBAL
--RBLFIS
--RBLMED
--RBLADD
--ABLADD
A P S A D D
W R ZF L X
Fig 7.1 Sebroucin* Mp for tht DIF3D Code Block (Cont'd)
97
(see Section 7*4.1) and computing the coefficients required for the calcula-
tion of peak power densities and peak fluxes in the edit overlay.
Solution edits in the nodal option are performed using many of the same
subroutines used in the finite difference option. Several subroutines have
been added to the edit overlay (DSSTOU) to accommodate the nodal option.
HEXBAL computes the leakage integrals in hexagonal geometry, and WNHFLX writes
the interface file NHFLUX used to restart nodal calculations. The subroutines
NHS HAP, NHPEAK, and NHPKED are used in computing the peak power densities and
peak total fluxes in the nodal option.
7.2 Data Management Strategy
The data management strategy in the DIF3D nodal option is similar to that
in the finite difference option. In the following discussion, the generic
terms FCM (fast cere memory) and ECM (extended core memory) refer to the two
different container arrays employed by the BFO1NTER package
29
in DIF3D. Both
containers are in fast-core memory on one-level computers such as the IBM 370
series. On two-level computers, of which the CDC 7600 is the only example at
present, the FCM container is in SCM (small-core memory) and the ECM container
is in LCM (large-core memory). As in the finite difference options,
3
core
files are allocated in the ECM container, and are therefore referred to as
ECM files. The discussion in this section pertains to the allocation of ECM
files and the transfer of data between ECM and disk files. The additional
transfer of data between ECM and FCM required on the CDC 7600 computer is
discussed in Section 7.3.
As mentioned in Section 6.1.1, the data management strategy is determined
by the amount of ECM storage allocated on A.DIF3D card type 02. The minimum
ECM storage requirement is given by the storage necessary to core-contain all
group-independent files plus the storage for one-group ECM buffers for the
group-dependent files stored on disk. Available storage beyond this minimum
requirement is used first to ECM-contain the scattering band of fluxes, and
then to contain as many of the group-dependent files as possible. This
strategy is implemented in the subroutine NHCORE. Since all data (i.e. all
mesh planes) for at least one group must be in ECM, there is no motivation to
implement an algorithm analogous to the concurrent inner iteration strategy
(CIIS) employed in the finite difference option.
The calculation of the peak power densities and peak fluxes in the nodal
option requires several large files (FLXSHP, PWDSHP, and PEAKNH) that are not
required during the outer iteration procedure. As a result, the storage
required for the edit overlay may exceed that required for the outer
iterations. For this reason, separate data management strategies are used
for the eigenvalue calculation and the edit overlay. This situation is
illustrated by the sample data management page shown in Fig. 6.1. The minimum
storage requirement (all data for 1 group in ECM during the outer iterations)
is calculated as described in the previous paragraph, subject to the additional
constraint that the minimum amount of storage required by the edit overlay also
be available* Once this minimum requirement is satisfied, an attempt is made
to core-contain the scattering band of fluxes followed by the group-dependent
files (flux-monents, new hex-plane partial currents, etc) in the order shown
in Fig* 6*1. In determining the location of files during the edit overlay,
98
all required group-dependent files (with the exception of the cross section
file, which retains its location during the outer iterations) are assumed to
be on disk initially with 1-group ECM buffers. These files are then trans-
ferred to core in the order shown in Fig. 6.1 as storage permits. Note that
it is possible for the partial current and flux moment files to be in core
during the outer iterations and on disk during the edit overlay. In this
case the subroutine CPYFIL (called from NHEDDM) is used to copy the file to
the proper location for the edit overlay.
The scratch disk files used in the nodal option are listed in Table 7.1.
Note that most of these files (e.g. FLXSHP, CCOEF, etc.) replace finite
difference files which are not used in the nodal calculation. Data are
transferred between these disk files and the corresponding ECM files shown
in Table 7.2 via calls to the DIF3D data management routines
3
BLKGET and
BLKPUT, which in turn call the standardized subroutines
29
*
30
DRED and DRIT
specified by the Committee on Computer Code Coordination. Files for which no
disk name is given in Table 7.2 are group-independent and are always contained
in ECM. The remaining group-dependent ECM files are opened (via calls to
OPENCF) with enough words to hold data for all groups if the file is core-
contained. If the file is stored on disk, the corresponding ECM file is
opened for only one group, and this space is used to buffer data as it is
written one group at a time to and from the disk file. If the scattering band
of fluxes is core-contained, the ECM file FLUX is opened with enough words to
hold data for MAXSCT+1 groups, where MAXSCT is the maximum number of down-
scatter groups.
Table 7.1 Scratch Disk Files Used in the Nodal Option
File
Group
Number
2
3
5
1
2
3
4
5
4
1
6
4
2
3
File
Reference
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
SEEK Table
File Name
RNDM01
RNDM02
RNDM03
RNDM04
RNDM05
RNDM06
RNDM07
RNDM08
RNDM09
RNDM10
RNDM11
RNDM12
RNDM13
RNDM14
Finite
Difference
File Name
PSIOLD
PSINEW
PSIUP
FDCOEF
FRNOLD
FRNNEW
FRNM1
FRNM2
SRCNEW
Z ONMAP
CXSECT
FSRC
PSIGO
PSIGN
Nodal
File Name
PSIOLD
PSINEW
FLXSHP
CCOEF
PCHOLD
PCHNEW
PCZ OLD
PCZ NEW

CXSECT

FLXOLD
FLXNEW
99
Table 7.2 Correspondence Between ECM and Disk Files in the Nodal Option
ECM Disk
File Name File Name File Contents
NHMAP
IPCPNT
IHXPNT
ICCPNT
CMR
XYLEAK
FSRC
FSRCi
FLUX
PCURRH
PCURRZ
PCURH1
PCURZ1
CXSECT
CCOEF
Composition Map in Nodal Ordering
Pointers to Hex-Plane Partial Currents
Pointers to Mesh-Cells in Nodal Ordering
Pointers to Nodal Coupling Coefficients
Storage for Coarse-Mesh Rebalance Matrix
Axial Moments of Hex-Plane Leakage
Fission Source Moments
FSRC From Previous Outer Iteration
Nodal Flux Moments
Hex-Plane Partial Currents
Axial Partial Currents
PCURRH From Previous Outer Iteration
PCURRZ From Previous Outer Iteration
Cross Sections
Nodal Coupling Coefficients
VOLUME
ZONMAP
PSINEW
FLXSHP
PWDSHP
PEAKNH
PSINEW,
PSIOLD
FLXSHP
Mesh-Cell Volumes in GEODST Ordering
Composition Map in GEODST Ordering
Nodal Fluxes in GEODST Ordering
Flux Shape Coefficients Used to Compute
Peak Values in Edit Overlay
Power-Density i^nape Coefficients
Nodal Peak Values by Mesh Cell
The ECM files listed in the upper section of Table 7.2 are defined (via
calls to DEFICF) in subroutines NHINIT and NHGORE, while those files shown in
the lower half are defined in subroutine NHEDDM and are used only in the DIF3D
edit overlay. (The files VOLUME, PSINEW, and FLXSHP are calculated in the
secondary overlay DNHFIN shown in Fig. 7.1). All disk files used in the nodal
option are defined in subroutine NHDISK using calls to DEFIDF. The character-
istics of the files shown in Table 7.2 are given by the calling arguments in
DEFICF and DEFIDF, and thus will not be given here. Smaller arrays not shown
in Table 7.2 are stored in the FCM container array via calls to the BPOINTER
routine PUTM, and are described in the comment cards included in the sub-
routines in which they are used*
100
7.3 Additional Data Management Considerations for the CDC 7600 Computer
The CCCC subroutines
29
*
30
CRED and CRIT are used to transfer data
between ECM and FCM on two-level computers such as the CDC 7600.* The use of
these routines thus avoids direct addressing of extended core memory. The
two-level Implementation of the nodal option requires that one-group data for
a single mesh plane plus all cross sections and coupling coefficients for a
single group fit in the available fast core memory. The finite difference
option also requires that all cross sections for a single group fit in FCM,
but permits blocking on the mesh plane such that only pointwise data for one
or more mesh lines need be stored in FCM. Full-core models of large fast
reactor designs with at least 200 zones (material compositions) can be
accommodated by the nodal option on the CDC 7600 computer.
The nodal two-level data management strategy is implemented in the
following manner. The FCM array SCRFCM is used to store all data which is
transferred between ECM and FCM. The pointers to the scratch sub-arrays held
in SCRFCM are not fixed throughout the calculation, but are instead recalcu-
lated at various stages of the calculation in accordance with the FCM storage
required at these points. The length of this array is thus determined as the
maximum scratch FCM storage required at any point in the nodal calculation.
This dynamic allocation procedure minimizes the size of the SCRFCM array since
only data actually used at each stage of the calculation is stored in SCRFCM.
In order to further clarify this procedure, we consider for example the
FCM storage required in subroutine 0UTR4, which calls FLXHEX, FLXZ , FSUPDT,
and CMMTRX (see Fig. 7.1). These latter subroutines, like all of the lowest-
level nodal subroutines, process data for a single mesh plane at a time.
Therefore, SCRFCM must contain all single-plane data used by the subroutines
called by OUTR4. This situation is illustrated in Table 7.3. A total of 10
scratch arrays (e.g. IANPNT, IZ NMAP, etc.) are used by FLXHEX, FLXZ , FSUPDT,
and CMMTRX as these subroutines are called in order from within a loop over
mesh planes. Data for the arrays IANPNT, IZ NMAP, FLUX, and CXSECT are read
for each plane from ECM into consecutive storage locations in SCRFCM as shown
in Table 7.3. These arrays retain their locations in SCRFCM during the four
calls required for each mesh plane. The arrays.PCH and IPCPNT are read into
SCRFCM and used during the execution of FLXHEX, but this same space is then
used by the arrays XYLEAK, PCZ , and FSRC during execution of FLXZ and FSUPDT.
CMMTRX then requires PCH, IPCPNT, and PCZ again, so these arrays are once
again read into SCRFCM. The asterisk in front of these arrays denotes the
fact that these arrays must be read prior to the call to CMMTRX. Arrays which
are recalculated (e.g. the FLUX array in this example) are then written from
SCRFCM back to ECM.
Tables such as that shown in Table 7.3 appear in the comment cards of the
source code wherever this dynamic allocation of SCRFCM is used. The reader is
reminded that this procedure is employed only in the two-level implementation
*Recall that the generic terms FCM and ECM refer to SCM (small core memory)
and LCM (large core memory) on the CDC 7600 computer.
101
of the 0IF3D nodal option. However, the SCRFCM array is used in the one-level
implementation to store two scratch arrays (QSRC and BMATRX) used in OUTR3 and
OUTR5.
Table 7.3 Allocation of Scratch Arrays in the FCM Array SCRFCM During
Subroutine OUTR4 (Two-Level Implementation Only)
Subroutine FLXHEX FLXZ FSUPDT CMMTRX
(1)
(2)
(3)
(4)
(5)
(6)
IANPNT
IZ NMAP
FLUX
CXSECT
PCH
IPCPNT (7)
(8)
IANPNT
IANMAP
FLUX
CXSECT
XYLEAK
PCZ
IANPNT
IANMAP
FLUX
CXSECT
(9) FSRC
< *)
(*)
(*)
(10)
IANPNT
IANMAP
FLUX
CXSECT
PCH
IPCPNT
PCZ
CMR
Before discussing the subroutines used to transfer data between ECM and
FCM, it is necessary to discuss briefly the structure of the ECM files shown
in Table 7.2. The ECM files are blocked into records such that each record
consists of one-group data for a single mesh plane. Two exceptions to this
rule are the cross section and coupling coefficient files in which a record
contains all data for a single energy group. These latter files thus contain
NGROUP records, while, for example, the group-dependent ECM flux file contains
a total of NGR0UP*KM records, where KM is the number of axial planes. The
nodal two-level data management strategy thus involves the transfer of complete
records of data between the ECM files and the SCRFCM array.
Data is transferred between ECM files and SCRFCM via calls to the utility
routines PCRED, PCRIT, ICRED, and ICRIT. These routines, which are not used
at present in the finite difference option, transfer a single record between
an ECM file and SCRFCM on two-level computers using calls to the CCCC routines
CRED and CRIT. They are also called in the one-level .implementation, where
they simply return the array pointer relative to the ECM container, which is
stored in fast core memory on one-level computers. These functions are
explained further in the comment cards appearing in the listing of subroutine
PCRED.
The single exception to the above procedure occurs in the treatment of
the group-independent file PWDSHP in the DIF3D edit overlay. This array
involves a total of 32 words per node in three-dimensional calculations. Due
to the large amount of data stored for each plane, data in this array are
transferred between ECM and FCM in blocks which may correspond to some frac-
tion of the nodes on the plane. The length of the sub-block is determined in
NHCORE.
102
7*4 Additional Programming Details
7.4.1 Mesh-Cell Ordering
The two different mesh cell orderings used in the nodal option are
illustrated in Fig. 7.2 for a two-dimensional sixth-core model with 6 rings of
hexagons. The GEODST ordering refers to the ordering of hexagons employed in
the CCCC geometry interface file GEODST (Ref. 30). This file is written by
the Input processor GNIP4C and subsequently processed by subroutine RGEODS in
the DIF3D code block. In the GEODST ordering, the hexagons are numbered by
rows, with "background" hexagons added to fill out the mesh such that all rows
have the same number of hexagons. The second ordering shown in Fig. 7.2 is
used during the nodal calculation controlled by the primary overlay NHSST.
The hexagons are ordered starting with the central hexagon and then moving
outward in a counterclockwise spiral. This ordering includes hexagons along
only one of the two symmetry lines [the excluded hexagons are treated via
period., (rotational symmetry) boundary conditions], and does not require any
background hexagons if all rings are full (as is the case in Fig. 7.2). The
nodal ordering thus reduces storage requirements, and is better suited to the
type of calculations (partial current sweeps, coarse-mesh rebalance on rings
of hexagons) performed in the nodal option. The GEODST ordering is used in
all interface files (e.g. RTFLUX, PWDINT, PKEDIT, etc.) written by the nodal
option, with the exception of the nodal restart file NHFLUX, which is written
in nodal ordering.
GEODST ORDERING
NODAL ORDERING
Fig. 7.2 GEODST and Nodal Mesh-Cell Orderings (Sixth-Core Symmetry)
103
The ECM file IHXPNT includes three separate records containing informa-
tion related to the ordering of the mesh cells* The first record, an array
IANPNT, contains pointers to "active" nodes, i.e. nodes which are included in
the actual reactor model as opposed to the background nodes used to fill in
the broken outer rings typical of LMFBR designs. Use of the pointers in
IANPNT permits sweeps over active nodes only, without use of IF testing in the
loop over mesh cells. The second record in IHXPNT is ICBPNT, which contains
pointers to the active nodes in the four-color checkerboard ordering used in
computing the hex-plane partial currents. The final record in IHXPNT is
ITRMAP, a transformation map between the nodal and GEODST orderings. ITRMAP
is used to reorder fluxes computed in the nodal overlay into the GEODST
ordering used in the DIF3D edit overlay. IANPNT and ICBPNT are calculated in
subroutine NHPNT, and ITRMAP is computed in HEXMAP. Since all mesh planes in
the three-dimensional mesh must have the same outer hex-plane boundary, it is
necessary to store these pointers only for a single plane.
7.4.2 Partial Current Ordering
The hex-plane and z-directed partial currents are stored in the ECM
files PCURRH and PCURRZ , respectively. PCURRH contains NGROUP*KM records,
where KM is the number of axial mesh planes. Each record contains a total
of NPCHEX (= 6*NHEX + NPCBDY) partial currents, where NHEX is the number of
hexagons on a plane (including background nodes) and NPCBDY is the number of
incoming partial currents on the outer hex-plane boundary. The six outgoing
partial currents are stored consecutively for each node as shown in Eq. (2.82)
and the nodes are ordered in the nodal ordering illustrated in Fig. 7.2. The
6*NHEX outgoing partial currents for the plane are followed by the NPCBDY
incoming partial currents on the boundary. The file PCURRZ contains
NGROUP*(KM+1) records, where KM+1 is the number of axial mesh boundaries.
Each record consists of 2*NHEX partial currents: the NHEX partial currents
in the negative z-direction plus the NHEX partial currents in the positive
z-direction.
The efficient execution of the algorithms presented in Section 4 requires
the use of pre-commited pointers to the six incoming hex-plane partial currents
for each node. These incoming partial currents are either outgoing partial
currents from neighboring nodes, incoming partial currents on the outer hex-
plane boundary, or, in the case of fractional-core models (e.g. Fig. 7.2) with
periodic boundary conditions, outgoing partial currents across surfaces along
a periodic boundary. The pointers are stored in the ECM file ICCPNT, which
consists of a single record of length 6*NHEX. For example, if
IPIN-IPCPNT(J.K), then PCURRH(IPIN) is the incoming partial current on hex-
plane surface J (Jl,...,6) for node K and the current energy group. A
different set of pointers (stored in the FCM array IPCBDY) are used to compute
incoming partial currents on the outer hex-plane boundary.
7.4.3 Storage of the Nodal Coupling Coefficients
As discussed in Section 3*5, nodal coupling coefficients are
stored only for unique nodes characterized by material composition (CCCC zone)
assignment and axial mesh spacing* The number of unique nodes is determined
104
in NHCORE via calls to NHCCPT. Pointers to unique nodes are stored in the
ECM file ICCPNT. For example, if NTYP - ICCPNT(K), then (CCOEF(I,NTYP),
I-l,...,13) contains the 13 coupling coefficients for the K-th node and the
current energy group in three-dimensional calculations. The 13 unique
coefficients are stored in the order shown in Eq. (B.32). In two-dimensional
calculations, only 7 unique coefficients are required, and they are stored in
the order shown in Eq. ( A.51) .
105
ACKNOWLEDGEMENTS
The author is indebted to Keith Derstine for his valuable assistance in
the implementation of the nodal scheme in the D1F3D code, and for his critical
review of the manuscript. Numerous discussions with Cy Adams, Greg Greenman,
and Herb Henryson II are gratefully acknowledged* Bert Toppel provided the
REBUS-3 calculations for the LCCEWG benchmark problem analyzed in Section 5.
The superb efforts of Marge Ramirez in typing this report are greatly
appreciated.
106
REFERENCES
1. R. W. Hardie, and W. W. Little, Jr., "3DB, A Three-Dimensional Diffusion
Theory Burnup Code," BNWL-1204, Battelle-Pacific Northwest Laboratories,
1970.
2. D. R. Vondy, T. B. Fowler, and G. W. Cunningham, "VENTURE: A Code Block
for Solving Multigroup Neutronics Problems Applying the Finite-Difference
Diffusion-Theory Approximation to Neutron Transport," ORNL-5062, Oak
Ridge National Laboratory, 1975.
3. K. L. Derstine, "DIF3D: A Code to Solve One-, Two-, and Three-Dimensional
Finite-Difference Diffusion Theory Problems," ANL-82-64, Argonne National
Laboratory, 1983.
4. C. H. Adams, "SYN3D: A Single Channel Spatial Flux Synthesis Code for
Diffusion Theory Calculations," ANL-76-21, Argonne National Laboratory,
1976.
5. B. J. Toppel, "A User's Guide for the REBUS-3 Fuel Cycle Analysis
Capability," ANL-83-2, Argonne National Laboratory, 1983.
6. A. F. Henry, "Refinements in Accuracy of Coarse-Mesh Finite-Difference
Solution of the Group Diffusion Equations," in Numerical Reactor Calcula-
tions, p. 447, International Atomic Energy Agency, Vienna, 1972.
7. See for example: J. D o mi n g , "Modern Coarse-Mesh Methods - A Development
of the '70's," Proceedings of the Topical Meeting on Computational
Methods in Nuclear Engineering, Williamsburg, VA., Vrl. 1, p. 3-2,
American Nuclear Society, April, 1979.
8. J. J. Arkuszewski, and M. Makai, "Analytical Coarse Mesh Approximations
for Solving Diffusion Equations in Hexagonal and Square Geometries," Pro-
ceedings of the International Topical Meeting on Advances in Mathematical
Methods for the Solution of Nuclear Engineering Problems, Munich, Vol. 2,
p. 89, American Nuclear Society/European Nuclear Society, April, 1981.
9. T. Duracz, "A Nodal Method in Hexagonal Geometry," Proceedings of the
International Topical Meeting on Advances in Mathematical Methods for
the Solution of Nuclear Engineering Problems, Munich, Vol. 1, p. 423,
American Nuclear Society/European Nuclear Society, April, 1981.
10. R. D. Lawrence, "A Nodal Interface Current Method for Multigroup Diffusion
Calculations in Hexagonal Geometry," Trans* Am. Nucl. So c , 39, 461 (1981).
11. R. D. Lawrence, "A Nodal Method for Three-Dimensional Fast Reactor
Calculations in Hexagonal Geometry," Proceedings of the Topical Meeting
on Advances in Reactor Computations, Vol. II, p. 1030, Salt Lake City,
American Nuclear Society, March, 1983.
12* Allan F. Henry, Nuclear-Reactor Analysis, MIT Press, Cambridge, Mass*
(1975).
107
13. R. D. Lawrence, "A Nodal Method for the Solution of the Neutron Diffusion
Equation in Hexagonal Geometry," Intra-Laboratory Memorandum, Argonne
National Laboratory, February, 1981*
14. D. R. Ferguson and K. L. Deretine, "Optimized Iteration Strategies and
Data Management Considerations for Fast Reactor Finite Difference Diffu-
sion Theory Codes," Nucl. Sci. Engr., 6 4 , 593 (1977).
15. H. Finnemann, F. Bennewitz, and M. R. Wagner, "Interface Current
Techniques for Multidimensional Reactor Calculations," Atomkernenergie,
3 0 , 123 (1977).
16. R. N. Sims and A. F. Henry, "A Coarse-Mesh Nodal Diffusion Method Based
on Response Matrix Considerations," MITNE-197, Massachusetts Institute
of Technology, 1977.
17. C. Maeder, "A Nodal Diffusion Method with Legendre Polynomials," P r o -
ceedings, Advances in Reactor Physics, Gatlinburg, Tennessee, p. 131,
American Nuclear Society, April 1978.
18. G. Greenman, K. Smith, and A. F. Henry, "Recent Advances in an Analytic
Nodal Method for Static and Transient Reactor Analysis," Proceedings of
the Topical Meeting on Computational Methods in Nuclear Engineering,
Williamsburg, VA., p. 3-49, American Nuclear Society, April, 1979.
19. R. A. Shober, "A Nodal Method for Fast Reactor Analysis," Proceedings of
the Topical Meeting on Computational Methods jLn Nuclear Engineering,
Williamsburg, VA., p. 3-33, American Nuclear Society, April, 1979.
20. R. D. Lawrence and J. J. D o mi n g , "A Nodal Green's Function Method for
Multidimensional Neutron Diffusion Calculations," Nucl. Sci. Engr., 7 6 ,
218 (1980).
21. E . L. Wachpress, Iterative Solution of Elliptic Systems, Prentice-Hall,
Inc., Englewood Cliffs, N.J. (1966).
22. R. Froehlich, "A Theoretical Foundation for Coarse-Mesh Variational
Techniques," GA-7870, Gulf General Atomic, 1968.
23. M. R. Wagner, "GAUGE - A Two-Dimensional Few Group Neutron Diffusion -
Depletion Program for a Uniform Triangular Mesh," GA-8307, Gulf General
Atomic, 1968.
24. G. Buckel, K. Kufner, and B. Stehle, "Benchmark Calculations for a Sodium-
Cooled Breeder Reactor by Two- and Three-Dimensional Diffusion Methods,"
Nucl. Sci. Engr. 64, 75 (1977).
25. "Benchmark Problem Book," ANL-7416, Supplement 3, Argonne National
Laboratory, to appear, 1983.
26. J. W . Lewellen, private communication, 1982.
108
27 . B. J. Toppel, C* H. Adams, R. D. Lawrence, H. Henryson II, and K. L.
Derstine, "Validation of Alternative Methods and Data for a Benchmark
Fast Reactor Depletion Calculation," Proceedings of the Topical Meeting
on Advances In Reactor Physics and Core Thermal Hydraulics, (held in
Kiamesha Lake, NY ) , NUREG/CP-0034, Vol. 1, p. 177, U.S. Nuclear
Regulatory Commission, Washington, D.C., August 1982.
28. B. J. Toppel, private communication, 1982.
29. C. H. Adams, K. L. Derstine, H. Henryson II, R. P. Hosteny and B. J.
Toppel, The Utility Subroutine Package Used by_ Applied Physics Export
Codes, ANL-83-3, Argonne National Laboratory, 1983.
30. R. Douglas O'Dell, "Standard Interface Files and Procedures for Reactor
Physics Codes, Version IV," LA-6941-MS, Los Alamos National Laboratory,
1977.
109
APPENDIX A
DERIVATION OF THE TWO-DIMENSIONAL RESPONSE MATRIX EQUATION
The purpose of this Appendix is to provide additional details of
the steps leading from the expression [Eq. (2.80)] for the interface
partial current to the final form [Eq. (2.85)] of the two-dimensional
response matrix equation. Before doing so, however, it is necessary to
give some additional details concerning the calculation of the expansion
coefficients a|
x
3 and the term EJS
x
(h/2) which appear in Eq. (2.80).
A.I Calculation of the Expansion Coefficient a
g X
3
The following expression [Eq. (2.41)] for a^ , was derived in
Section 2.4.1:
g X J
a
k
= E
k
gx3 " g>
where
L *
s
^(x,-y (x)) - 2 ^ (x). (A.2)
gX 5 e g o g *
The final form of the equation for 3g
X
3
w a s
then derived by
(1) developing a relationship [Eq. (2.43)] between E
X
( X ) and the
leakage term *|y(x) defined in Eq. (2.42), (2) approximating *|y(x)
as shown in Eq. (2.44), and (3) combining these results to obtain a
final equation [Eq. (2.53)] for a]|
x
3. These steps are discussed in
order in the following subsections.
A. 1.1 The Relationship Between Eg
X
(x) and JEg
V
(x)
As discussed in Section (2.4.1), the form of Eq. (A.2) suggests
that EJ|
x
(x) can be related to the y-directed leakage defined by
( A
-
3 )
no
This relationship is derived by expanding < j> |(x,y) in a Taylor series
around the point ( x , 0 ) :
ay
y-0
2
ay y-o
z
y3
3
-
ay
y-o
(A. 4 )
Evaluating Eq. (A.4) at y y
s
( x ) and then summing the results yields
ay
y-o
(A. 5 )
Using Eq. ( A.4 ) ,
- k , v
d> ( x) = 9
g x
i r
s J
-
<
2
ay
y-0
+ (Xh
1
*). (A. 6)
Using Eq. (A.6) to eliminate < j> H(x,O) in Eq. (A.5) yields
y-0
(A. 7)
Since
2
ay' I y-0
D
8 y
g
Ill
Eq. (A*7) becomes
Cx)]
f (
s gy
g
This result was given as Eq. (2.43) in Section 2.4.3.
A*1.2 Approximation of
g
y(x)
The y-directed leakage is approximated by Eq. (2.44):
f_ -h/2 < x < 0

k
gy+
0 < x < h/2
where *gy- and gy+ are half-node averages defined in Eqs. (2.45).
The initial steps in the derivation leading to expressions for these
averages are given by Eqs. (2.46) through (2.51). We continue this
derivation here by substituting Eq. (2.49) into Eq. (2.50b) to yield
- 2 J U
-k
i i - n/ z j - J K-n.it) - - . T** <- s- i o i x . v I X J J t - d> i x . - v I X J J I
-h/2
2 r
h / 2
Ta k
/ <** feO*.y)| " ) (A. io) lyy _ (x);
From Eqs. (A.2) and (2.51b):
,y Cx)) + 4^(x,-y (x)) - E^ (x) + 2 ^ (x) (A. 11)
s s B gx gx
112
h/2
y - y
H
U ) l k
k
V T
Substituting Eqs. (A.11) and (A.12) into Eq. (A.10) yields

y +
(A. 13)
Neglecting the O(h**) error in Eq. (A.8):
(h/2) ^

e v +
( A
*
1 4 )
g
E
k
(0) = 4 [E
k
< 0") + E
k
gx
N
' 2 gx g
g
Substituting Eqs. (A.14) and (A.15) into Eq. (A.13) and simplifying
yields
(+h/2) - J
k
(-h/2) - tl7 f
k
- 2
k
]
gu ' gv
v
18 gy+ gy-
J
2D
k
^
l
*g
X
( + h / 2 )
" *g x
( 0 ) 1
* (A.16a)
Substituting Eq. (2.49) into Eq. (2.50a), using Eq. (2.51a), and
then following the above procedure leads to the analogous equation
113
[17 f
k
- 2
2D
k
?L (h /2) ))]. (A.16b)
Q gX gX
Solving for Eqs. (A.16) for f
k
and
k
+
yields
*
k
" T ^ (r U
k
(+h/2) - J
k
(-h/2)] + -Tj& [^ (+h/2) - I
k
(0)]
gy- 95
l
h gu
v
' gv
v
h
2
v v
g x ' *gx
ir
[ J
l
( + h / 2 )
"
5
L
(
"
h / 2 ) 1 +
~ 2

i*L
(
"
h/ 2)
- *L
(O)1
1-
(A
-
17a)
34D
k
f
3
4D
k
( A
'
1 7 b)
A.1.3 Final Form of the Equation for
Substituting Eqs . (A. 17) into Eq. (2.52) yields
k 2 t?
w
fc
M
1C 4 fe. **lc ^lt
g
where
L
k
= J
k
(-Hi/2) - J
k
(-h/2)
gu g u gu
^ v =
5
g v
( + h / 2 )
" ^ v
(
114
Using Eq. (2 .33),
If
Substituting Eq. (A.13) Into Eq. (A.18), and then solving for a
yields
TW JT
|E
gu
+
fy IT Sj < * >
+
*
g
Th is final result for a was g iven as Eq. (2 .53) in Section 2 .4.3.
I,
A.2 Calculation of Eg
X
(h / 2 )
jf_
An ex pression for E
g X
(h / 2 ) is obtained by substituting Eq.
(A.17b) into Eq. (A.14), using Eq. (A.19) to eliminate g
X
(0),
and th en simplifying th e result:
E
g
X
( h / 2 ) =
" i k ^
s
TI35 t
1 7 9
* g x
( + h / 2 ) + 4 9
*gx^
h/2
> "
2 2 8
*gl
+
T30
a
A.3 Th e Response Matrix Equation
An ex pression for th e outg oing surfaceaverag ed partial current
across th e surface in ch e positive x direction was g iven as Eq. (2 .80):
D
k
D
J
OU t
'
k
( +h / 2 ) & [
a
k
t
+ if
a
k
+
7
a
k 1 k k
g x h g x l 13 g x 2 2 6 g x 3 2 g x 4 g x
jin, k
( + h / 2 ) # ( A > 2 2 )
It is convenient tc introduce constants 0, and &2 (wh ich are eith er
aero or one) in order to disting uish th e approx imations obtained for
N2 ,3, or 4 in Eq. (2 .33), i.e.
115
D
k
- - i {
a
k
+
36
gx h
l
gxl 13
. 1
7 h V ^
+ J
g
( + h / 2 )

(A,23)
where
N2 Sj M , e
2
= o
N-3 Pj s 1, 3
2
s 0 (A.24)
N-4 *
Note that consistent with Eq. (A.I), E.
x
(x) does not enter into the
calculation for N2 in Eq. (2.33).
The derivation of the final form of the response matrix equation
follows the five steps described iri Section 2.6. We repeat these steps
in more detail nere:
(1) Using Eqs. (2.34a) and (2.34b) yields
Combining Eqs. (A.20) and (A.21) yields
7 If V 129 9 -'if 7S'l
16 V 3
+ E
gx<
h
/
2
' ' " W ^
+1
-/
2
> " 1157
2052 -
k
4 h
r
=-
k
, -
k
,
5187 *g " T 0 5 ^
l L
g u
+
gv
J
g
i+n/^^ - J v-n/zjj. (A.26)
OJ _K, gV gU
g
'>
116
Using Eq. (2.59b) yields
* - -
12
J
+
! l^<*/
2
> "
T *g*4 -
1 2
Jgx!
+
! l^c<*/
2
> g. Ch/a)]. (A.27)
Substituting Eqs. (A.25), (A.26), and (A.27) into Eq. (A.23) and
collecting terms yields
^g 49 1299 8 -k
gx " " h I
[
l 3 " 3 1 8 7
P
l
+
3
e
2
J <
l
)
g
X
<
+h
/
2
>
v
k
^ J
J( h / 2) ] + J^'
k
(+h/2) (A.28)
k
(2) The nodal fl ux <| > i s obtained from Eq. ( 2. 74 ) :
+ L
k
+ L
k
] , (A.29)
8U 8 V
J L ? ^ [ L L L
8
hS
r

k 8
3hZ
r

k 8 X 8U 8 V
g S
k k
whi l e the fl ux moment <t>
eX
i s at i s fi e s Eq. ( 2. 70) [with a
e x
i evaluated
using Eq. (2. 34 a. )]:
S
L-
Q
k
2 ^k
+
^k . j k j
k ^gxl ^k gx gu gv
J
D
k
* ^ <
+ h
/
2
> ^ ( h / 2 ) J (A.30)
H7
where
gl g h
Substituting Eqs. (A.29) and (A.30) into Eq. (A.28) and collecting
terms yields
-W> - - {i x
x
^ I ^ ^ i* ^
t E
gu
+ E
g v
] + T
5 <
J
g u
( + h / 2 )
'
T
7
T
8 Qg
+ T
9
Q
gxl
where
2D
k
2D
T
- I >
f
72 _ 2052 ,
T
3 * J
K
g 0
l
13 5187
B
1
J
(A. 33c)
<A
'
3 3 d
>
155
( A
-
33f )
118
H 8 K* So (A.33g)
8
1
*
(A.33h)
(A.331)
and
k
D

/ h
^O'-VT
(A
'
34a)
gu
h 2
r , K
g
k
D
!>
K . 5 - . (A. 34b)
g i
a ,
gl
(3) The surface-averaged fluxes, net currents, and leakages on the right
hand side of Eq. (A.32) are eliminated in favor of partial currents using
the relationships
k
(h/2) - 2 [J
O u t , k
(h /2) + J
i n , k
(h /2)] (A.35a)
gx gx gx
J
k
(+h/2) J
o u t

k
( +h / 2) - J
n

k
( +h /2) (A.35b)
gx gx gx
J
k
(-h/2) - J^'
k
( - h / 2) - J"
t
'
k
(-h/2) (A.35c)
gx gx gx.
^
v
f ( / ) ^ ( / ) f ( / ) ^ ( / ) (A.35d)
gx gx gx gx gx
j
( 4 i l
/ 2 ) - j
i n

k
( +h /2) - J
0 U t , k
( - h / 2) + J
i n , k
( - h / 2) . (A.35e)
gx gx gx gx gx
119
Analogous expressions are used to eliminate the corresponding u - and
v-direction t e r ms . Substituting Eqs . ( A.35) into Eq. (A.32} and then
collecting terms yields
a,,] J
u t

k
,,] J b
2
]
,
T
in,k
c
2
c
3
c
4
c
3
c
2
] J
g
(A.36)
where J * and J * are column vectors containing the six
-g -g
outgoing and six incoming partial currents, respectively, for the
k-th node, e.g.
J
O U t

k
= col [ J
Ou t

k
( +h / 2) , J
OU t

k
( +h / 2) , J
o u t
'
k
( +h / 2) , J
U t
'
k
( - h / 2) ,
~g gx gu gv gx
(A. 37)
The constants introduced in Eq. (A.36) are defined as follows:
= 1 + Tj + T
3
+ (A.38a)
(A.38b)
(A.38c)
(A.38d)
b
l
H T
8
(A.39*)
(A. 39b)
120
C
2
S a
2
C
4
5
"
T
2
+ T
3 "
T
7
(A.40a)
(A.40b)
(A.40c)
(A.40d)
(4) Five additional equations similar to Eq* (A*36) can be obtained
by applying successive 60 rotational transformations to Eq. (A.36).
These equations can be combined with Eq. (A.36) to yield
[A
k
] J
OU t
'
k
g -8
where
[B
k
]
8
[C
k
J J
l
8 " 8
(A.41)
S (A.42)
a
l
a
2
a
3
a
4
a
3
a
2
a
2
a
l
a
2
a
3
a
4
a
3
a
3
a
2
a
l
a
2
a
3
a
4
a
4
a
3
a
2
a
l
a
2
a
3
a
3
a
2
a
l
a
2
a
2
a
3
a
4
a
3
a
2
a
l
(A. 43)
b, 0
'
S
0
0
"
b
2
0
0
0
0
-b
fl
0
0
b
2
0
0
- b .
(A.44)
and
121
c
l
C
2
C
3
C
4
C
3
C
2
C
2
C
l
C
2
C
3
C
4
C
3
C
3
C
2
C
l
C
2
C
3
C
4
C
4
C
3
C
2
c
l
C
2
C
3
C
2
c
l
k k
Note that [A_] and [C
g
] are symmetric matrices with structures
consistent with the 60 transformations used to generate the u - and
v-direction analogs of Eq. (A.36). The negative entries in [ B | ]
appear since Q|
x
i is an odd spatial moment over the interval
xe[-h/2,+h/2].
(A.45)
(5) The final form of the response matrix equation is obtained by
inverting [AJ|] in Eq. (A.41) to yield
out.k
8 8
{
kj jin,k
8 -8 *
(A.46)
w h e r e
[ P
g
J S 3 _1 [ B
g
J
(A.47)
(A.48)
k k
The structures of the response matrix [R
g
] and the source matrix [P
g
]
can be deduced* from symmetry considerations:
l R
g
] S
120
'180
120
-60
120
180
120
120
u
60
"60
120
180
180
'120
L
60
120
120
180
120
L
60
60
60
120
180
120
u
60
(A. 49 )
122
'0
Pi
p
2
p
2
P2
p
l
p
2
"P2
p
2
p
l
"
p
"Pi
P
0
P
2
~P
2
(A.50)
The entries in these matrices have the following physical interpre-
tations:
r = reflection coefficient
= transmission coefficient between two surfaces oriented at 60'
t,~
n
= transmission coefficient between two surfaces oriented at 120
t.
g0
= transmission coefficient between two surfaces oriented at 180
P
n
= zero-moment source coefficient specifying contribution to an
outgoing partial current from the node-averaged group source
term
p. = first-moment source coefficient specifying contribution to an
outgoing partial current in a given direction from the first-
moment group source term in the same direction
p
?
= first-moment source coefficient specifying contribution to an
outgoing partial current in a given direction from the first-
moment group source term in a different direction.
Note that these coefficients are defined for a single energy group g, and
include only effects due to in-group diffusion and removal* Group to group
transfer due to fission and in-scatter are included in the source term g^
appearing in Eq. (A.46). The entries of [R|] and [ P |] are computed in
8
the following manner* Let m^ denote the first row of the inverse of the
[A|] matrix defined in Eq. (A.43). Comparison of Eqs. (A.48) and (A.47)
with Eqs. (A.49) and (A.50) yields
123
r - m
T
[c, c c, c, c c J
T
(A.51a)
T T
t
60
-
S
t c
2
c
l
C
2
C
3
C
4
C
3
]
(A.51b)
^20
=
S t
c
3
c
2
c
l
C
2
C
3
C
4
]
(A.51c)
T T
t
180
=
- *
C
4
C
3
C
2
c
l
C
2
C
3 ^ (A.51d)
T T
P
o
* m [bj bj bj bj b
t
bj] (A.51e)
a
T
[b
2
0 0 - b
2
0 0 ]
T
(A.51f)
p
2
- m
T
[0 b
2
0 0 - b
2
0 ]
T
. , (A.51g)
Inspection of Eqs. (A.33) shows that the 7 coefficients given in Eqs.
(A.51) depend only upon the diffusion coefficient DJ and the removal
cross section 2g> k in the k-th node and g-th energy group. Hence
these coefficients are computed and stored only for unique nodes
characterized by their material composition assignment.
12 4
APPEND IX B
D ERIVATION OF THE THREED IMENSIONAL RESPONSE MATRIX EQUATION
In th is Appendix we provide additional details of th e steps leading
from th e ex pressions [Eqs* (3.50) and (3*51)] for th e interface partial
currents to th e final form [Eq. (3.58)] of th e th reedimensional response
matrix equation*
As in Section A.3 of Appendix A, Eq. (3.50) is rewritten in th e form
J *
wh ere th e constants 3, and &
2
a r e
defined in Eq. (A.2 4). Similarly,
an additional constant 3, is introduced in order to disting uish th e
approx imations obtained for N
z
* 2 ,3 in Eq. (3.2 9). Th e ex pression
[Eq. (3.51)] for th e outg oing surfaceaverag ed partial current across
th e surface in th e positive zdirection is th en
D
k
g
k
" " " ' ' i Q _ T J n TT P ^ n j T J I TuZ f J % \ Om J
wh ere
N 2 * g, i 0
Z J
(B.3)
N 3 =*> e
0
5 l .
z 3
Th e derivation of th e final form of th e th reedimensional response
matrix equation follows th e five steps described in Section 3.5. We
repeat th ese steps in more detail h ere, making use of th e results derived
in Appendix A*
(1) Th e elimination of th e ex pansion coefficients in Eq. (B.I) is dis
cussed in Section A.3. Th e result is identical to th e twodimensional
result g iven in Eq. (A.2 8).
Th e elimination of th e ex pansion coefficients in Eq. (B.2 ) proceeds
in th e following manner* Using Eqs. (3.30a) and (3.30b) yields:
125
a
k
, + 3a
k
o
- 4d>
k
(+Az
k
/ 2) + 2$>
k
( - Az
k
/ 2) - 6*
k
. (B.4)
gzx gz/ gz gz g
Using Eq. (3.37) yields:
4 - -60
8 l
+ 5[^
z
(+A
2
k
/2) - *
k
z
(-Az
k
/2)]. (B.5)
Substituting Eqs. (B.4) and (B.5) into Eq. (B.2) and collecting terms
yields
D
k
{[4 + 50J *
k
(+Az
k
/2)
J gZ
[2 - 5fS
3
] ;
k
z
(-Az/) *
k k
while the flux moments <j>
gx
j and <j
z
i satisfy Eqs. (3.39) and (3.44)
[with a|
x
i and a|
2
i evaluated using Eqs. (2.34a) and (3.30a)]:
A.
3 a
D
k
9a -
,=k . -k _ =k ,
IT
gx
+ T
gu V
3
(B.6)
(2) The nodal flux
k
is obtained from Eq. (3.38):
O
~ [L
K
+ L
K
+ L
K
] , , L
K
, (B.7)
^r . k gx gu gv
J
k r,k gz
yo
"'
g g g
126
L . ro
k
- 1 -
k
8zl 3h
i - A. 5
gxyzl
J
2 k _r,k gz
D
k
-Sj- ~ [f
z
(-ttz /2) - $* (-Az /2)J. (B.9)
8
Substitution of Eqs. (B.7) and (B.8) into Eq. (A.28) and collecting
terms yields
T
i n
L
k
+ T . [L
k
+ L
k
10 gz 4 gu gv
x
c
[J
k
(+h/2) - J
k
(-
5 gu gv
x, [J
k
(+h/2) - J
k
(-h/2)] + x
7
[f
k
+ f
k
- f
k
]
6 gv
N
g u
x
7 gx gu gv
J
where T , through T Q are defined in Eqs. (A.33), and
Substituting Eqs. (B.7) and (B.9) into Eq. (B.6) and collecting terms
yields
127
T
z6
where
2D
k
T
z5 =
30 K
gz0
and
g
2D
k
T
z2
E
- f i
2
" I
5 + 60 K
gz0
] B
3l <
B
'
13b
>
Az
T
z3
E 6 K
gz0
(B.13f)
(B.13g)
<">
128
(3) The surface-averaged fluxes, net currents, and leakages on the right
hand sides of Eqs. (B.10) and (B.12) are eliminated in favor of partial
currents using the relationships shown in Eqs. (A.35) and their u-, v-,
and z-direction analogs. Equations (B.10) and (B.11) can then be written
as
1
.
a
2
a
3
a
4
a
3
a
2
a
5
a
5
]

5
1
.out,k
m
j i n. k
-8
b
l
b
2
]
.^gxl
1
. k
Az
L
k
gzxl .
(B.15)
and
r
1
T
OUt , k
r
, , ,
[a
6
a
6
a
6
a
G
a
6
a
6
a
7
a
8
]
4
= [b
3
b
4
3
2_
L
k
3h gxyz1
[c
i
T
i n , k
c
?
c
8
] J
g
, (B.16)
respectively, where J
u
* and J ' are column vectors containing
the eight outgoing ant eight incoming partial currents, respectively,
for the k-th node, e.g.
J
o u t
>
k
= col [J
Ou t
'
k
( +h /2) , J
OU t
'
k
( +h / 2\ J
Ou t

k
( +b /2) , J
OU t
'
k
( - h / 2) ,

g gx gu gv gx
Those constants introduced in Eqs. (B.15) and (B.16) that are not given
in Eqs. (A..38) through (A.40) are defined as follows:
a
5
H T
10
(B.18a)
6 z4
(B.18b)
129
a
7
H X + T
z l
+ T
z 3
+ T
z 5
(B.18c)
(B.18d)
(4) Five additional equations similar to Eq. (B.15) can be obtained
by applying successive 60 rotational transformations to Eq* (B.15).
Calculation of
Jj^
,
g (-Az
k
/2) yields an additional equation similar
to Eq. (B.16). These additional equations can be combined with Eqs.
(B.15) and (B.16) to yield
where
(B.19a)
(B.19b)
C
7 =
+ T
z 3
+ T
z 5
C
8
E
"
T
z 2
+ T
Z
3
(B.20a)
(B.20b)
( B
'
20 c
>
( B
'
2 0 d )
"

c al
130
'
;
a
l
a
2
a
3
a
4
a
3
a
2
a
5
a
5
a
2
a
l
a
2
a
3
a
4
a
3
a
5
a
5
a
3
a
2
a
l
a
2
a
3
a
4
a
5
a
5
a
3
a
4
a
3
a
2
a
l
a
2
a
5
a
5
a
6
a
6
a
6
a
6
a
6
a
6
a
7
a
8 .
a
6
a
6
a
6
a
6
a
6
a
6
a
8
a
7
(B.24)
[ B
g
] S
b
2
0
0
"
b
2
0
0
0
0
0
b
2
0
0
"
b
2
0
0
0
0
0
b
2
0
0
"
b
2
0
0
0
0
0
0
0
0
\
- b
(B.25)
and
131
]
c
l
C
2
C
3
C
4
C
3
C
2
C
5
C
5
C
2
c
l
C
2
C
3
C
4
C
3
C
5
C
5
C
3
C
2
c
l
C
2
C
3
C
4
C
5
C
5
q
2
C
3
C
4
C
3
C
2
c
l
C
5
C
5
C
6
C
6
C
6
C
6
C
6
C
6
C
7
C
8
C
6
C
6
C
6
C
6
C
6
C
6
C
8
C
7
(B.26)
(5) The final form of the three-dimensional response matrix equation
Is obtained by inverting [A^] in Eq. (B.21) to yield
out.k
m k ]
i k _ k^
(
k j ln.k
-g g
J ia
g "g
J
g ~g *
(B.27)
where
(B.28)
(B.29)
k k
The structures of the response matrix [R
g
] and the source matrix [P
g
J
can be deduced from symmetry considerations:
132
,k, _
]
r
'60
fc
120
* 180
t
1 20
'60
xy +z
c
9 0
xy +z
c
9 0
'60
r
*60
C
1 20
C
1 8 0
t
1 20
xy +z
c
9 0
x y +z
c
9 0
t
1 20
fc
60
r
*60
t
1 20
t
1 8 0
x y +z
c
9 0
xy +z
c
9 0
' 1 8 0
C
1 20
'60
r
'60
'l2d
xy*z
c
9 0
xy+z
90
fc
120
'180
t
1 20 .
'60
r
'60
xy+z
C
9 0
xy+z
c
9 0
C
60 '
C
1 20
t
1 8 0
t
1 20
'60
r
xy+z
c
9 0
xy*z
c
9 0
z+xy
c
9 0
z+xy
C
9 0
z+xy
c
9 0
z+xy
c
9 0
z+xy
C
9 0
z+xy
C
9 0
z
fc
180
z+xy
c
9 0
z+xy
c
9 0
z+xy
c
9 0
z+xy
c
9 0
z+xy
Z
90
z+xy
c
9 0 '
'180
z
[R
[R
gxy
k
[R
gxyz
[R
k
g
1 [R
gzxy gz
P
O
P
l
P
O
P
2
P
O "
P
2
P
0
_ P
1
p
O "
P
2
P
O
P
2
p
2
Pl
p
2
_p
2
"Pi
"
p
2
"
p
2
p
2
p
l
P
2
-
p
2
"
p
l
0
0
0
0
0
0
0
0
(B.30)
( B .31)
133
t.
The additional entries in the three-dimensional forms of [R
g
] and
[Fg] have the following physical interpretations:
t g
0
= transmission coefficient from an axial surface to a hex-
plane surface
xy^z
*"90
5
transmission coefficient from a hex-plane surface to an
axial surface
9
r = reflection coefficient for axial surfaces
z
= transmission coefficient between axial surfaces
P Q = axial zero-moment source coefficient specifying contri-
bution to an outgoing partial current on an axial surface
from the node-averaged group source term.
p, = axial first-moment source coefficient specifying contri-
bution to an outgoing partial current on an axial surface
from the axial first-moment group source term.
k k
The entries of [R
g
] and [Pg] are computed in the following manner.
Let m? and ml denote the first and seventh rows of the inverse of the
k -1 "'
[Ag] matrix defined in Eq. (B.24). Comparison of Eqs. (B.29) and (B.28)
with Eqs. (B.30) and (B.31) yields
T T
r
" Si f
c
i
C
2
C
3
C
4
C
3
C
2
C
5
C
5^ (B.32a)
fc
60 " Si
I c
2
c
l
C
2
C
3
C
4
C
3
C
5
C
5
1
(B.32b)
T
C
2
c
l
C
2
C
3
C
4
C
5
C
5
3
(B.32c)
t
180 " Si
fc
4
C
3
C
2
c
l
C
2
C
3
C
5
c
5
i T
(B.32d)
C
5
C
5
C
5
C
5
C
5
C
7
C
8
] T ( B
-
3 2 e )
134
S7 Icj c
2
c
3
c
4
c
3
c
2
c
6
c
6
]
T
(B.32f)
r
z " 27 *
C
5
C
5
C
5
C
5
C
5
C
5
C
7
C
8
J
(B.32g)
" 27
Ic
5
C
5
C
5
C
5
C
5
C
5
C
8
C
7
, T
(B.32h)
P
0
" Si t
b
i
b
i
b
i
b
X
b
x
b
!
b
3
b
3
I
T
(B.321)
p
l " - I
Ib
2 "
b
2
0]T
(B.32J)
p
2
- m [ [0 b
2
0 0 - b
2
0 0 0 ]
T
(B.32k)
Q " 27 I
b
!
b
i
b
i
b
i
b
i
b
i
b
3
b
3
l
T
(B.321)
j - m^ [0 0 0 0 0 0 b
4
- b
4
]
T
. (B.32m)
Thus, in three dimensions, a total of 13 coefficients are computed and
stored for each unique node characterized by its material composition
assignment and axial mesh spacing.
135
APPENDIX C
APPROXIMATION OF THE LEAKAGE MOMENTS
In this Appendix we provide additional details concerning the
approximation of the leakage moments which appear in the three-dimensional
response matrix equation, and then summarize results for a model problem
study designed to isolate the errors associated with these approximations.
C.I Calculation of the Leakage Moments
As discussed in Section 3.6, the leakage moments are calculated
using the approximations
L
g z
( x
'
y ) E
gz
L
g x y
( z )
~
p
g x y
( z )
'
z e A z k
'
< C
'
2 )
where
p
gxy
( z ) E E
gxy
+ p
gxyi
f
* i
( z ) + p
gxy2
f
z2<
z
>'
k k
The expansion coefficients Pg
xy
i and P
g X
y 2
a r e
calculated using the
constraints given in Eqs. (3.66). The results are
k
i = 4r T - { [2Az
k
" + Az
k
] [Az
k_
+ Az
k
] [f r
+
- L
k
]
gxyl 3h d
l J
gxy gxy
J
Az
k+
] [Az
k
+ 2Az
k+
] [L
k
- L
k
" ]} (C.4a)
k _ 2Az
k
V
10
,
r
. k - , . k ,
r
rk+
f
k
p
gxy2 = I T d" *
l Az + Lz ] [ L
gxy "
L
Az
k +
] [ L ^ - L
k
;
y
] } , (C.4b)
136
where d is defined in Eq. (3.69c). Substituting Eq. (C.Aa) into Eq. (3.67)
yields the final result for L&xyl shown in Eq. (3.68).
For nodes adjacent to an outer reactor boundary (with vacuum boundary
conditions), the quadratic leakage approximation is constructed such that
( c
*
5 )
where & denotes a node adjacent to an outer axial boundary such that the
local coordinate Az^/2 denotes the nodal surface which forms part of this
boundary. Equation (C.5) is used since, for a zero flux boundary con-
dition, the transverse gradient of the flux evaluated on the surface is
identically zero. For the case in which a zero incoming partial current
boundary condition is imposed, the transverse gradient is not zero; hence
Eq. (C.5) is only approximate, although we expect this approximation to
be sufficiently accurate for systems in which the extrapolation length is
small. The quadratic expansion coefficients for these surface nodes can
be calculated using Eqs. (C.4); the boundary condition given in Eq. (C.5)
is satisfied by setting to zero the a?ial mesh spacing (e.g. Az^
+
) and
the average hex-plane leakage (e.g. Lg
X
y) f
r
"nodes" lying outside
the outer reactor boundary*
C.2 Model Problem Study of the Errors Associated with the Leakage
Approximations
A simple homogeneous model problem was analyzed in order to assess
the effect of the approximations introduced in Eqs. (C.I) and (C.2) on
the overall accuracy of the three-dimensional nodal scheme. The model
problem consists of a central hexagon surrounded by two full rings of
hexagons, with an axial height of 200 cm and a lattice pitch of 20 cm.
Z ero flux boundary conditions are imposed on all external surfaces.
Two-group cross sections typical of a light water reactor were used in
the calculations; these cross sections are given in Table C.I.
Table C.1 Two-Group Cross Sections for the Model Problem
Group
1
2
g
D
1
0
8
. 5
. 4
0.
0.
g
0023
2
0
0
g
. 0
. 218
z
s
0.06

X
1
0
g
. 0
. 0
Since the solution to this homogeneous problem is separable in the
hex-plane and axial directions, i.e.
137
*
g
( x,y ,z)
*x y
( x

y )
z
- I t
*,
the following exact expressions for the leakage moments are obtained:
J
gzxl
(C.6)
gxyzl
~ u J
"gzl
(C.7)
Thus the differences in the solutions computed using these exact results
and the approximations given in Eqs. (C.I) and (C.2) provide a measure of
the errors introduced by the leakage approximations.
Table C.2 shows the results of eigenvalue calculations using the
indicated approximations to the leakages. All nodal calculations were
done using a 20 cm axial mesh spacing. The eigenvalue for calculation
number 5 was obtained using Richardson extrapolation of DIF3D finite
difference results, and is thus assumed to be the exact solution to the
diffusion equation for this problem.
Table C.2 Computed Eigenvalues for the Model Problem
Calculation
Number
1
2
3
4
Approximation
to L (x,y)
Flat
Flat
Flat
Exact
5 (Extrapolated finite
Approximation
to L g
x y
( z )
Flat
Quadratic
Exact
Quadratic
difference solution)
Eigenvalue
(k-eff)
0.96989
0.96935
0.96934
0.96907
0.96842
138
Table C.3 lists the eigenvalue errors associated with these leakage
approximations* The bases for these error estimates are also shown. Note
that a flat approximation to L|
x
y(z) [obtained with Pgxyl =
p
gxy2 = 0 in
Eq. (C.3)] introduces an error or 0.00055, but this error is essentially
eliminated by the use of the quadratic approximation. The total error in
the reference nodal calculation [i.e. using Eqs. (C.I) and (C.2)] is
0.00096 relative to the extrapolated finite difference solution. This
ertor thus includes contributions (0.00001 and 0.00028, respectively) due
to the quadratic approximation of Lg
X
y(
z
) and the flat approximation
of L|
z
(x,y). The difference (0.00067) between the total error and these
leakage contributions is due to the remaining hex-plane approximations
[specifically, Eqs. (2.33), (2.43), and (2.44)]. (Results not included in
in Table C.2 have shown that the error due to the axial polynomial approxi-
mation [Eq. (3.29)] is negligable). This difference is consistent with
results obtained for the two-dimensional version of this problem.
Table C.3 Error Contributions for the Model Problem
Source of Error Basis
Error
Contribution
Flat Approximation to L (z)
gxy
Quadratic Approximation to L (z)
gxy
Flat Approximation to L (x,y)
gz
All Approximations in Reference
Nodal Calculation
k
l
k
2
k
2
k
2
"
k
3
- k
3
~
k
4
"
k
5
+ 0.00055
+ 0.00001
+ 0.00028
+ 0.00096
e.g. ki is the eigenvalue computed in calculation number 1
of Table C.2.
D
The reference nodal calculation is calculation number 2 of
Table C.2.
The results of this model problem study suggest the following
conclusions:
(1) The error associated with the quadratic approximation to
is very small relative to the overall error in the nodal calculation. This
quadratic approximation leads to a dramatic improvement in accuracy relative
to the flat approximation to L |
x v
( z ) .
139
(2) The flat approximation to LJ|
z
(x,y) introduces a significant
contribution (0.00028) to the overall error (0.00096) in the nodal calcu-
lation. The results confirming the accuracy of the quadratic approximation
to L g x y (
z
) suggest that an analogous approximation (discussed in Section 3.6)
to l g
Z
( x,y ) could essentially eliminate this error.
140
APPENDIX D
CALCULATION OF PEAK POWER DENSITIES
One drawback of the nodal approach is the lack of information
concerning the spatial distribution of the flux within the node* This
information is required during the edit overlay in order to compute peak
power densities and fluxes in each node* A simple procedure is used in
the nodal option to compute more accurate peak power densities and fluxes
than those obtained by sampling the available node-averaged values* In
two dimensions, this procedure involves sampling surface-averaged values
on the six surfaces of the k-th node; the surface fluxes are readily
obtained from the available interface partial currents:
* g ^ ) - 2 [J
U t

k
(S
1
) + J^' ^Sj) ] , i - 1 6. (D.I)
The notation introduced here differs from that used in Sections 2 and 3.
In particular, the six surfaces are numbered beginning with the surface
in the positive x-direction and then moving counterclockwise around the
hexagon. The surface-averaged values of the power density are
C
where PC is a power conversion factor. The peak power density in the
k-th node is
P
4t ,ma X
E max P
k
(S ) , i = 0.....6, (D
O
3)
i
where
P C
g K '
(D
'
4)
8=1
An analogous procedure (with PC
g
= 1) is used to compute the peak total
fluxes edited by DIF3D.
This procedure is extended to three dimensions by assuming the flux
within the node is separable in the hex-plane and axial directions:
141
where, consistent with the notation in Section 3*2,
-If 1 I t
' '--
N
- ' dz < j> (x,y,z) (D.6)
h/2 y
g
(x)
*
( z )
/
d x
/
d y
*
,,
^h/2 ^- y jx )
Using Eq. (D.5) the axial dependence of the hex-plane-averaged flux
on the six hex-plane surfaces is given by
*g
( S
i '
z )
" *g x y
( S
i
)
*gz
-k
where < j)g
X
y(
s
i)>
t h
e surface-averaged fluxes on the six hex-plane
surfaces, are computed as in Eq. (D.I). The axial shape function
is evaluated using Eq. (3.29):
N
z a
k
n-1 *g
k k k
The coefficients &azl *
a
gz2
an<
*
a
g z3
a r e
computed using Eqs.
(3.30a), (3.30b), and (3.37) respectively. Using Eq. (D.8),
142
8
8-1
The peak power density in the k-th node is computed by evaluating Eq.
(D.10) at J+l equally-spaced axial elevations
z. = -Az
k
/2 +4 ^z
k
J " 0,...,J, J = 10.
T h e pe a k de n s i t y i s t h u s
P
k

m a x
= ma x P
k
(S
i
, z ) , i - 0 , . . . , 6 , j - 0 J, (D . l l )
w h e r e
8
As shown by the results presented in Section 5.3, the accuracy of
this simple scheme is comparable to that of a fine-mesh finite difference
calculation. Sampling the surface-averaged values offers a distinct
advantage in two-dimensional fast reactor calculations since the peak
values in the inner and radial blankets occur at the core-blanket inter-
face. Adequate accuracy is obtained in the driver fuel regions due to
the relatively small dimensions of the hexagonal fuel assemblies. The
validity of the separability assumption used in extending this scheme
to three dimensions depends upon the degree of heterogeneity in the
axial direction. Although it is clear that this approximation will not
be very accurate in the vicinity of inserted control rods, it is likely
that this assumption will be valid in the region away from control rods
and blankets where the driver-fuel peak power density typically occurs.
If the computed peak-to-average value in a node is unrealistically high
(greater than 5), it is assumed that the separability approximation is
not valid, and the peak value in the node is computed by sampling only
the node- and surface-averaged values. This "fixup" has been required
only for nodes in control assemblies for the test problems studied to
date.
143
APPENDIX E
DESCRIPTION OF THE BCD INPUT FILE A.DIF3D
C***********************************************************************
C
C REVISED 1/20/82
C
CF A.DIF3D
CE ONE-, TWO-, AND THREE-DIMENSIONAL DIFFUSION THEORY
CE MODULE-DEPENDENT BCD INPUT
C
CN THIS BCD DATASET MAY BE WRITTEN EITHER
CN IN FREE FORMAT (UNFORM-A.DIF3D) OR
CN ACCORDING TO THE FORMATS SPECIFIED FOR EACH
CN CARD TYPE (DATASET-A.DIF3D).
CN
CN COLUMNS 1-2 MUST CONTAIN THE CARD TYPE NUMBER. -
CN
CN A BLANK OR ZERO FIELD GIVES THE DEFAULT OPTION -
CN INDICATED.
CN
CN NON-DEFAULTED DATA ITEMS ON THE A.DIF3D
CN DATA SET ALWAYS OVERRIDE THE CORRESPONDING
CN DATA ON THE RESTART DATA SET DIF3D.
CN ENTER -1 TO RESET DATA ON THE RESTART DATA SET -
CN DIF3D BACK TO THEIR DEFAULT VALUES.
C
C***********************************************************************
CR PROBLEM TITLE (TYPE 01)
C
CL FORMAT (I2,4X,11A6)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD 1-2 01
CD
CD 7-72 ANY ALPHANUMERIC CHARACTERS (1 CARD ONLY).
C
CR STORAGE AND DUMP SPECIFICATIONS (TYPE 02)
C
CL FORMAT (I2,4X,3I6)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
144
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
C
C
1-2 02
7-12 POINTR CONTAINER ARRAY SIZE IN FAST CORE MEMORY (FCM)
IN REAL*8 WORDS (DEFAULT-10000).
13-18 POINTR CONTAINER ARRAY SIZE IN EXTENDED CORE
MEMORY (ECM) IN REAL*8 WORDS (DEFAULT-30000).
19-24 POINTR DEBUGGING EDIT.
0...N0 DEBUGGING PRINTOUT (DEFAULT).
1...DEBUGGING DUMP PRINTOUT.
2...DEBUGGING TRACE PRINTOUT.
3...BOTH DUMP AND TRACE PRINTOUT.
C
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
PROBLEM CONTROL PARAMETERS (TYPE 03)
FORMAT (12,4X, 1116)
COLUMNS
1-2
7-12
13-18
CONTENTS...IMPLICATIONS, IF ANY
03
PROBLEM TYPE.
0...K-EFFECTIVE PROBLEM (DEFAULT).
1...FIXED SOURCE PROBLEM.
SOLUTION TYPE.
0...REAL SOLUTION (DEFAULT).
1...ADJOINT SOLUTION.
2...BOTH REAL AND ADJOINT SOLUTION.
19-24 CHEBYSHEV ACCELERATION OF OUTER ITERATIONS.
0...YES, ACCELERATE THE OUTER ITERATIONS (DEFAULT).
1...NO ACCELERATION.
25-30 MINIMUM PLANE-BLOCK (RECORD) SIZE IN REAL*8 WORDS FOR
I/O TRANSFER IN THE CONCURRENT INNER ITERATION
STRATEGY. THE DEFAULT (-4500) IS HIGHLY RECOMMENDED.
31-36 OUTER ITERATION CONTROL.
-3...BYPASS DIF3D MODULE.
-2...PERFORM NEUTRONICS EDITS ONLY.
-1...PERFORM NEUTRONICS EDITS AND CALCULATE OPTIMUM
OVERRELAXATION FACTORS ONLY.
.GE.O...MAXIMUM NUMBER OF OUTER ITERATIONS (DEFAULT-30).
37-42 RESTART FLAG.
0...THIS IS NOT A RESTART (DEFAULT).
145
CD 1...THIS IS A RESTART PROBLEM.
CD
CD 43-48 JOB TIME LIMIT, MAXIMUM (CP AND PP(OR WAIT)) PROCESSOR
CD SECONDS (DEFAULT-1000000000).
CD
CD 49-54 NUMBER OF UPSCATTER ITERATIONS PER OUTER ITERATION
CD (DEFAULT-5). PERTINENT TO UPSCATTER PROBLEMS ONLY.
CD
CD 55-60 CONCURRENT ITERATION EFFICIENCY OPTION.
CD 0...PERFORM THE ESTIMATED NO. OF INNER ITERATIONS FOR
CD EACH GROUP.
CD 1...AVOID THE LAST PASS OF INNER ITERATIONS IN THOSE
CD GROUPS FOR WHICH THE NO. OF ITERATIONS IN THE LAST
CD PASS ARE LESS THAN A CODE DEPENDENT THRESHOLD.
CD
CD 61-66 ACCELERATION OF OPTIMUM OVERRELAXATION FACTOR
CD CALCULATION.
CD 0...NO ACCELERATION (DEFAULT).
CD 1...ASYMPTOTIC SOURCE EXTRAPOLATION OF POWER ITERATIONS-
CD USED TO ESTIMATE THE SPECTRAL RADIUS OF EACH INNER
CD (WITHIN GROUP) ITERATION MATRIX.
CD 67-72 OPTIMUM OVERRELAXATION FACTOR ESTIMATION ITERATION
CD CONTROL. THE DEFAULT (-50) IS STRONGLY RECOMMENDED.
C
CN THE MAXIMUM NUMBER OF OUTER ITERATIONS SENTINEL
CN SPECIFIES THE NUMBER OF OUTERS THAT CAN BE PERFORMED
CN (COLS. 31-36) EACH TIME THE DIF3D MODULE IS INVOKED.
CN
CN THE DIF3D TERMINATION PROCEDURE WILL ALWAYS:
CN l...(RE)WRITE THE APPROPRIATE FLUX FILES
CN (RTFLUX OR AT F L U X) .
CN 2...(RE)WRITE THE RESTART FILE DIF3D.
CN TO FACILITATE AUTOMATIC RESTART, THE RESTART FLAG
CN ON THE DIF3D RESTART CONTROL FILE WILL BE TURNED ON
CN AUTOMATICALLY UPON DETECTION OF:
CN 1...MAXIMUM NUMBER OF OUTER ITERATIONS.
CN 2...TIME LIMIT.
CN
CN
CN TO RESTART THE FLUX CALCULATION:
CN EITHER
CN
CN PROVIDE THE RESTART DATA SET DIF3D AND
CN THE APPROPRIATE FLUX DATA SET (RTFLUX OR ATFLUX)
CN AND SPECIFY THEM UNDER "BLOCK-OLD" IN THE BCD
CN INPUT DATA
CN OR
CN 1...SET THE RESTART FLAG (COLS. 37-42) TO 1 ON
CN THE TYPE 03 CARD. THIS PERMITS IMMEDIATE
CN RESUMPTION OF OUTER ITERATION ACCELERATION.
CN 2...INCLUDE THE LATEST K-EFFECTIVE ESTIMATE
CN (COLS. 13-24) AND THE DOMINANCE RATIO
CN ESTIMATE ON THE TYPE 06 CARD (COLS. 61- 7 2) .
1A6
CN 3...INCLUDE THE OPTIMUM OVERRELAXATION FACTORS
CN FOR EACH GROUP (TYPE 07 C AR D ) .
CN 4...PROVIDE THE APPROPRIATE FLUX DATA SET (RTFLUX
CN OR ATFLUX) AND SPECIFY IT UNDER "BLOCK-OLD"
CN IN THE BCD INPUT DATA.
CN
CN A NON-Z ERO TIME LIMIT (COLS. 43-48) OVERRIDES
CN THE ACTUAL TIME LIMIT DETERMINED INTERNALLY
CN BY SYSTEM ROUTINES IN THE ANL AND LBL PRODUCTION
CN IMPLEMENTATIONS
CN
CN THE TIME LIMIT PARAMETER (COLS. 43-48) IS PERTINENT
CN TO EACH ENTRY TO THE DIF3D MODULE.
CN
CN IT IS RECOMMENDED THAT AN ODD NUMBER OF UPSCATTER
CN ITERATIONS BE SPECIFIED (COLS. 49-54) TO AVOID
CN ADDITIONAL I/O OVERHEAD.
CN
CN THE USER IS CAUTIONED TO MONITOR THE POINT-WISE
CN FISSION SOURCE CONVERGENCE TO ENSURE THAT MONOTONIC
CN CONVERGENCE IS OBTAINED WHEN THE EFFICIENCY OPTION
CN (COLS. 55-60) IS ACTIVATED.
CN
CN THE OPTIMUM OVERRELAXATION FACTOR ACCELERATION OPTION
CN IS PRIMARILY INTENDED FOR PROBLEMS KNOWN TO HAVE HIGH -
CN (> 1.8) OPTIMUM OVERRELAXATION FACTORS.
CN
CN ITERATION CONTROL (COLS. 67-72) OF THE OPTIMUM
CN OVERRELAXATION FACTOR ESTIMATION IS PRIMARILY INTENDED -
CN FOR USE IN CONJUNCTION WITH THE ASYMPTOTIC ACCELERATION-
CN OPTION (COLS. 61- 66) .
C
CR EDIT OPTIONS (TYPE 04)
C
CL FORMAT (12,4X,1016)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD ~ - ----.
CD 1-2 04
CD
CD 7-12 PROBLEM DESCRIPTION EDIT (IN ADDITION TO USER INPUT
CD SPECIFICATIONS WHICH ARE ALWAYS EDITED.
CD 0...N0 EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
C D 13-18 GEOMETRY (REGION TO MESH INTERVAL) MAP EDIT.
147
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 19-24 GEOMETRY (Z ONE TO MESH INTERVAL) MAP EDIT.
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 25-30 MACROSCOPIC CROSS SECTION EDIT.
CD ENTER TWO DIGIT NUMBER SP WHERE
CD
CD S CONTROLS THE SCATTERING AND PRINCIPAL CROSS SECTIONS -
CD P CONTROLS THE PRINCIPAL CROSS SECTIONS EDIT ONLY.
CD
CD THE INTEGERS S AND P SHOULD BE ASSIGNED ONE OF THE
CD FOLLOWING VALUES (LEADING Z EROES ARE IRRELEVANT).
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 31-36 BALANCE EDITS
CD ENTER 3 DIGIT NUMBER GBR WHERE
CD
CD G CONTROLS GROUP BALANCE EDITS INTEGRATED OVER THE
CD REACTOR
CD B CONTROLS REGION BALANCE EDIT BY GROUP
CD R CONTROLS REGION BALANCE EDIT TOTALS
CD (INCLUDING NET PRODUCTION AND ENERGY MEDIANS)
CD
CD THE INTEGERS G, B, AND R SHOULD BE ASSIGNED ONE OF THE -
CD FOLLOWING VALUES (LEADING Z EROES ARE IRRELEVANT)
CD 0...N0 EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 37-42 POWER EDITS
CD ENTER 2 DIGIT NUMBER RM WHERE
CD
CD R CONTROLS REGION POWER AND AVERAGE POWER DENSITY EDITS-
CD M CONTROLS POWER DENSITY BY MESH INTERVAL EDIT (PWDINT)-
CD
CD THE INTEGERS R AND M SHOULD BE ASSIGNED
CD ONE OF THE FOLLOWING VALUES (LEADING Z EROES ARE
CD IRRELEVANT)
CD O...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
148
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 43-48 FLUX EDITS
CD ENTER 3 DIGIT INTEGER RMB WHERE
CD
CD R CONTROLS FLUX EDIT BY REGION AND GROUP
CD INCLUDING GROUP AND REGION TOTALS
CD M CONTROLS TOTAL (GROUP INTEGRATED) FLUX EDIT
CD BY MESH INTERVAL
CD B CONTROLS TOTAL FLUX EDIT BY MESH INTERVAL AND GROUP -
CD (RTFLUX OR ATFLUX)
CD
CD THE INTEGERS R, M, AND B SHOULD BE ASSIGNED
CD ONE OF THE FOLLOWING VALUES (LEADING Z EROES ARE
CD IRRELEVANT)
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 49-54 Z ONE AVERAGED (REAL) FLUX EDIT.
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 55-60 REGION AVERAGED FLUX EDIT.
CD 0...NO EDITS (DEFAULT).
CD 1...PRINT EDITS.
CD 2...WRITE EDITS TO AUXILIARY OUTPUT FILE.
CD 3...WRITE EDITS TO BOTH PRINT AND AUXILIARY OUTPUT FILE-
CD
CD 61-66 STANDARD INTERFACE FILES TO BE WRITTEN IN ADDITION
CD TO RTFLUX AND/OR ATFLUX.
CD 0...NONE (DEFAULT).
CD 1...WRITE PWDINT.
CD 2...WRITE RZ FLUX.
CD 3...WRITE BOTH PWDINT AND RZ FLUX.
CD
CD 67-72 MASTER DIF3D EDIT SENTINEL DURING CRITICALITY SEARCHES -
C : -1...SUPPRESS ALL DIF3D EDITS EXCEPT THE ITERATION
CD HISTORY AND ERROR DIAGNOSTICS
CD O...EDIT INPUT DATA ON 1ST SEARCH PASS, OUTPUT
CD INTEGRALS UPON CONVERGENCE OR UPON ACHIEVING THE
CD MAXIMUM SEARCH PASS LIMIT.
CD N...ALSO INVOKE SPECIFIED DIF3D EDITS EVERY N-TH
C D SEARCH PASS.
C
CN MULTI-DIGIT EDIT SPECIFICATION EXAMPLES.
CN
CN ENTERING THE INTEGER 201 IN COLUMNS 31-36 YIELDS
C N THE GROUP BALANCE EDIT ON THE AUXILIARY FILE AND
149
GN
CN
CN
CN
CN
C
c-
THE REGION BALANCE EDIT ON THE PRIMARY PRINT FILE.
ENTERING THE INTEGER 30 IN COLUMNS 31-36 YIELDS
THE REGION BALANCE EDIT BY GROUP ON BOTH THE PRINT AND
THE AUXILIARY OUTPUT FILES.
C
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
C
CN
CN
CN
CN
C
C~
CONVERGENCE CRITERIA (TYPE 05)
F OR M AT (I2,10X,3E12.5) .
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
1-2 05
13-24 EIGENVALUE CONVERGENCE CRITERION FOR STEADY STATE
CALCULATION (DEFAULT VALUE - 1.0E-7 IS RECOMMENDED).
25-36 POINTWISE FISSION SOURCE CONVERGENCE CRITERION
FOR STEADY STATE SHAPE CALCULATION
(DEFAULT VALUE - 1.0E-5 IS RECOMMENDED).
37-48 AVERAGE FISSION SOURCE CONVERGENCE CRITERION
FOR STEADY STATE SHAPE CALCULATION
(DEFAULT VALUE - 1.0E-5 IS RECOMMENDED).
IN UPSCATTERING PROBLEMS IT IS RECOMMENDED THAT
THE EIGENVALUE CONVERGENCE CRITERION (COLS. 13-24)
BE .1 TIMES THE POINTWISE FISSION SOURCE CONVERGENCE
CRITERION (COLS. 25- 36) .
C
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
OTHER FLOATING POINT DATA (TYPE 06)
FORMAT (12,1OX, 5E12. 5)
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
1-2 06
13-24 K-EFFECTIVE OF REACTOR (DEFAULT IS OBTAINED FROM
THE APPROPRIATE RTFLUX OR ATFLUX FILE, IF PRESENT.
OTHERWISE DEFAULT - 1.0).
25-36 ANY POINTWISE FISSION SOURCE WILL BE NEGLECTED IN THE
POINTWISE FISSION SOURCE CONVERGENCE TEST IF IT IS
150
CD LESS THAN THIS FACTOR TIMES THE R.M.S. FISSION
CD SOURCE (DEFAULT VALUE .001 IS RECOMMENDED).
CD
CD 37-48 ERROR REDUCTION FACTOR TO BE ACHIEVED BY EACH SERIES
CD OF INNER ITERATIONS FOR EACH GROUP DURING A SHAPE
CD CALCULATION - STRONGLY RECOMMENDED THAT THE DEFAULT
CD VALUE OF (.04) BE USED.
CD
CD 49-60 STEADY STATE REACTOR POWER (WATTS). (DEFAULT - 1 . 0 ).
CD
CD 61-72 DOMINANCE RATIO (FOR RESTART JOBS ONLY).
C
CN K-EFFECTIVE SPECIFICATIONS (COLS. 13-24):
CN 1...FOR K-EFFECTIVE PROBLEMS, SUPPLY ESTIMATED
CN K-EFFECTIVE OF REACTOR.
CN 2...FOR RESTARTED K-EFFECTIVE PROBLEMS, SUPPLY
CN LATEST K-EFFECTIVE ESTIMATE SUPPLIED ON THE
CN ITERATION HISTORY EDIT.
CN 3...FOR SOURCE PROBLEMS, SUPPLY K-EFFECTIVE OF
CN THE REACTOR.
CN DEFAULT IS OBTAINED FROM THE APPROPRIATE RTFLUX OR
CN ATFLUX FILE, IF PRESENT. OTHERWISE DEFAULT-1.0 .
C
CN NONMONOTONIC POINTWISE FISSION SOURCE CONVERGENCE
CN IS USUALLY INDICATIVE OF THE NEED TO TIGHTEN THE ERROR
CN REDUCTION FACTOR(COLS. 37-48). THIS IS FREQUENTLY TRUE-
CN IN TRIANGULAR GEOMETRY PROBLEMS WHERE A VALUE OF .01 IS-
CN USUALLY SUFFICIENT TO OBTAIN MONOTONIC CONVERGENCE.
CR OPTIMUM OVERRELAXATION FACTORS (TYPE 07)
C
CL FORMAT (12,10X,5E12. 5)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD 1-2 07
CD
CD 13-24 OPTIMUM OVERRELAXATION FACTOR FOR GROUP 1.
CD
CD 25-36 OPTIMUM OVERRELAXATION FACTOR FOR GROUP 2.
CD
CD 37-48 OPTIMUM OVERRELAXATION FACTOR FOR GROUP 3.
CD
CD 49-60 OPTIMUM OVERRELAXATION FACTOR FOR GROUP 4.
CD
CD 61-72 OPTIMUM OVERRELAXATION FACTOR FOR GROUP 5.
C
CN REPEAT 5 VALUES PER CARD FOR AS MANY TYPE 07 CARDS
CN AS ARE NEEDED.
152
CN
CN
CN
C
PROBLEM EIGENVALUE WILL BE ESTIMATED. IN THIS CASE, IT
IS RECOMMENDED TO INCREASE THE NUMBER OF ITERATIONS IN
COLS. 7-12 TO AT LEAST 10.
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
C
CN
CN
CN
CN
SN TRANSPORT OPTIONS (TYPE 09)
FORMAT -(I2,4X,2I6,6X,E12.4)
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
1-2 09
7-12 SN ORDER.
13-18 MAXIMUM ALLOWED NUMBER OF LINE SWEEPS PER LINE PER
INNER ITERATION (DEFAULT-10).
25-36 LINE SWEEP CONVERGENCE CRITERION (DEFAULT-l.QE-4).
TO INVOKE THE DIF3D TRANSPORT OPTION, THE TYPE 09 CARD
MUST BE PRESENT WITH A NONZERO SN ORDER. FOR THE TIME
BEING, USERS MUST ALSO CONTINUE TO 'PRELIB
,
TO
DATASET 'C116.B99983.MODLIB
,
TO INVOKE THIS OPTION.
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
PARAMETERS FOR NODAL HEXAGONAL GEOMETRY OPTION (TYPE 10)
FORMAT (I2,4X,5I6)
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
1-2 10
7-12 ORDER OF NODAL APPROXIMATION IN HEX-PLANE.
2...NH2 APPROXIMATION.
3...NH3 APPROXIMATION.
4...NH4 APPROXIMATION (DEFAULT).
13-18 ORDER OF NODAL APPROXIMATION IN Z-DIRECTION.
2...QUADRATIC APPROXIMATION.
3...CUBIC APPROXIMATION (DEFAULT).
19-24 COARSE-MESH REBALANCE ACCELERATION CONTROL.
-1...NO COARSE-MESH REBALANCE ACCELERATION.
.GE.O...NUMBER OF COARSE-MESH REBALANCE ITERATIONS PER
OUTER ITERATION (DEFAULT-2).
151
CN
CN THE OPTIMUM OVERRELAXATION FACTORS ARE NORMALLY
CN OBTAINED FROM THE RESTART INSTRUCTIONS PRINTED
CN IMMEDIATELY AFTER THE DIF3D ITERATION HISTORY EDIT.
CN IN THE RESTART INSTRUCTIONS, THE FACTORS ARE ALWAYS
CN EDITTED IN THE R E A L PROBLEM ORDERING AND SHOULD BE
CN ENTERED ON THE TYPE 07 CARD E X A C T L Y AS EDITTED
CN IN THE RESTART INSTRUCTIONS.
CN
CN THE PERMISSIBLE FACTOR RANGE IS BOUNDED BY 1.0 AND 2.0
CN INCLUSIVE. A ZERO OR BLANK FACTOR ENTRY DEFAULTS
CN TO 1.0. FACTORS ARE COMPUTED FOR THOSE GROUPS HAVING
CN A FACTOR OF 1.0; FACTORS GREATER THAN 1.0 ARE NOT
CN RECOMPUTED.
CN
CN TYPE 07 CARDS ARE PRIMARILY INTENDED FOR RESTART JOBS
CN ONLY (STRONGLY RECOMMENDED).
C
CR NEAR CRITICAL SOURCE PROBLEM ASYMPTOTIC EXTRAPOLATION
CR PARAMETERS (TYPE 08)
C
CC ***** WARNING...SELECT THIS OPTION ONLY IF THE *****
CC ***** ASYMPTOTIC EXTRAPOLATION IS REQUIRED FOR *****
CC ***** THIS PROBLEM. *****
C
CL FORMAT (I2,4X,I6,E12. 5)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD 1-2 08
CD
CD 7-12 NUMBER OF OUTER (POWER) ITERATIONS PERFORMED PRIOR TO
CD ASYMPTOTIC EXTRAPOLATION OF NEAR CRITICAL SOURCE
CD PROBLEM (DEFAULT=5).
CD
CD 13-24 EIGENVALUE OF THE HOMOGENEOUS PROBLEM CORRESPONDING
CD TO THE NEAR CRITICAL SOURCE PROBLEM. THIS EIGENVALUE
CD MUST BE LESS THAN ONE.
CD
CD 25-30 INITIAL FLUX GUESS SENTINEL.
CD 0...FLAT FLUX GUESS-1.0 (DEFAULT)
CD 1...FLAT FLUX GUESS-0.0
C
CN THE TYPE 08 CARD IS REQUIRED TO ACTIVATE AN ALTERNATE
CN SPECIAL ACCELERATION SCHEME FOR NEAR CRITICAL
CN SOURCE PROBLEMS.
CN
CN IF COLS. 13-24 ARE ZERO OR BLANK, THE HOMOGENEOUS
153
CD
CD
CD
CD
CD
CD
CD
C
CN
CN
CN
CN
CN
CN
CN
CN
CN
C
c~
25-30 ASYMPTOTIC SOURCE EXTRAPOLATION OF OUTER ITERATIONS.
0...APPLY ASYMPTOTIC SOURCE EXTRAPOLATION TO OUTER
ITERATIONS (DEFAULT).
1...N0 ASYMPTOTIC SOURCE EXTRAPOLATION.
31-36 NUMBER OF AXIAL PARTIAL CURRENT SWEEPS PER GROUP
PER OUTER ITERATION (DEFAULT-2).
THE TYPE 10 CARD IS PERTINENT ONLY WHEN THE NODAL
HEXAGONAL GEOMETRY OPTION (A.NIP3 TYPE 03 CARD
GEOMETRY-TYPE SENTINEL VALUES BETWEEN 110 AND 128)
IS SPECIFIED.
IT IS RECOMMENDED THAT THE DEFAULT VALUES FOR THE
ORDER OF THE NODAL APPROXIMATION IN THE HEX-PLANE
(COLS. 7-12) AND FOR THE ORDER OF THE NODAL APPROXI-
MATION IN THE Z -DIRECTION (COLS. 13-18) BE SPECIFIED.
c
CR
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
C
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
AXIAL COARSE-MESH REBALANCE BOUNDARIES FOR NODAL
HEXAGONAL GEOMETRY OPTION (TYPE 11)
FORMAT (I2,10X,3(I6,E12.5))
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
1-2 11
13-18 NUMBER OF AXIAL COARSE-MESH REBALANCE INTERVALS.
19-30 UPPER Z -COORDINATE.
31-36 NUMBER OF AXIAL COARSE-MESH REBALANCE INTERVALS.
37-48 UPPER Z -COORDINATE.
49-54 NUMBER OF AXIAL COARSE-MESH REBALANCE INTERVALS.
55-66 UPPER Z -COORDINATE.
THE TYPE 11 CARD IS PERTINENT ONLY WHEN THE THREE-
DIMENSIONAL NODAL HEXAGONAL GEOMETRY OPTION (A.NIP3
TYPE 03 CARD GEOMETRY-TYPE SENTINEL VALUES BETWEEN
120 AND 128) IS SPECIFIED.
IF NO TYPE 11 CARDS ARE PRESENT, THE AXIAL COARSE-MESH
REBALANCE INTERVALS ARE DEFINED BY THE Z -COORDINATE
VALUES SPECIFIED ON A.NIP3 CARD 09.
BOUNDARIES ARE SPECIFIED VIA NUMBER PAIRS.
154
CN EACH NUMBER PAIR IS OF THE FORM (N(I), Z ( I) ) . THERE
CN ARE N(I) AXIAL COARSE-MESH REBALANCE INTERVALS BETWEEN
CN Z (I-l) AND Z ( I) , WHERE Z ( 0) IS THE LOWER REACTOR
CN BOUNDARY IN THE Z -DIRECTION. NUMBER PAIRS MUST BE
CN GIVEN IN ORDER OF INCREASING MESH COORDINATES. ALL
CN AXIAL COARSE-MESH REBALANCE BOUNDARIES MUST COINCIDE
CN WITH THE MESH LINES WHICH BOUND MESH INTERVALS.
CEOF
155
APPENDIX F
DESCRIPTION OF SELECTED CARD TYPES IN THE BCD INPUT FILE A.NIP3
C**********************************************************************-
PREPARED 8/28/75 AT ANL
LAST REVISED 03/25/82
A.NIP3
NEUTRONICS MODEL INPUT FOR CODES WHICH REQUIRE CCCC
INTERFACE FILES
THIS BCD DATA SET MAY BE WRITTEN EITHER
IN FREE FORMAT (UNF0RM=A.NIP3) OR ACCORDING TO -
THE FORMATS SPECIFIED FOR EACH CARD TYPE
(DATASET=A.NIP3).
COLUMNS 1-2 MUST CONTAIN THE CARD TYPE
NUMBER.
UNLESS OTHERWISE STATED, BLANKS ARE NOT
MEANINGFUL IN A6 LABEL FIELDS.
*** CARD TYPE DIRECTORY ***
CONTENTS
PROBLEM TITLE
INPUT PROCESSING SPECIFICATIONS
PROBLEM GEOMETRY
EXTERNAL BOUNDARY CONDITIONS
EXTERNAL BOUNDARY CONDITION CONSTANTS
REGION BOUNDARIES FOR ORTHOGONAL GEOMETRIES
AREA SPECIFICATIONS
VARIABLE-MESH STRUCTURE
INTERNAL BLACK ABSORBER CONDITIONS
INTERNAL BLACK ABSORBER CONDITION CONSTANTS
FINITE-GEOMETRY TRANSVERSE DISTANCES
MATERIAL SPECIFICATIONS
COMPOSITION (ZONE) SPECIFICATIONS
REGION/COMPOSITION CORRESPONDENCE
REGION OR MESH DISTRIBUTED INHOMOGENEOUS SOURCE
SEARCH EDIT OPTIONS AND CONVERGENCE CRITERIA
SEARCH PARAMETER DATA
CONCENTRATION MODIFIERS FOR CRITICALITY SEARCH
MESH MODIFIERS FOR CRITICALITY SEARCH
BUCKLING MODIFIERS FOR CRITICALITY SEARCH
ALPHA MODIFIERS FOR CRITICALITY SEARCH
HEXAGON DIMENSION
REGION DEFINITIONS FOR ARRAYS OF HEXAGONS
c
c
c
c
CF
CE
CE
C
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
C
C
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
TYP
01
02
03
04
05
06
07
09
10
11
12
13
14
15
19
21
22
23
24
25
26
29
30
156
BACKGROUND REGION FOR ARRAYS OF HEXAGONS
COMPOSITION- AND GROUP-DEPENDENT BUCKLINGS
DIRECTIONAL DIFFUSION COEF. SCHEME
DIRECTIONAL DIFFUSION COEF./COMPOSITION CORRESPONDENCE
FISSION ENERGY CONVERSION FACTORS
CAPTURE ENERGY CONVERSION FACTORS
NUCLIDE SET ASSIGNMENTS
SOURCE EDIT, SYNTHESIS TRIAL FUNCTION SOURCE
NATURAL DECAY INHOMOGENEOUS SOURCE
SOURCE SPECTRA
GRAPHICS OUTPUT CONTROL
c********************************************************^
CR PROBLEM GEOMETRY SPECIFICATION (TYPE 03)
C
CL F O R M A T (1 2, 1 0 X , 16)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY.
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
CN
C
31
34
35
36
37
38
39
40
41
42
43
CD 1-2 03
CD
CD 13-18 GEOMETRY TYPE.
CD 10...SLAB
CD 20...CYLINDER
CD 30...SPHERE
CD 40...X-Y
CD 44...X-Y-Z
CD 50...R-Z
CD 60...R-THETA
CD 62...R-THETA-Z
CD 64...THETA-R
CD 66...THETA-R-Z
CD 70...TRIANGULAR, RHOMBIC BOUNDARY, CORE CENTER AT
CD 60 DEGREE ANGLE (SIXTH CORE SYMMETRY).
CD 72...TRIANGULAR, RECTANGULAR BOUNDARY, HALF CORE
CD SYMMETRY.
CD 74...TRIANGULAR, RHOMBIC BOUNDARY, CORE CENTER AT
CD 120 DEGREE ANGLE (THIRD CORE SYMMETRY).
CD 76...TRIANGULAR, 60 DEGREE TRIANGULAR BOUNDARY,
CD SIXTH CORE SYMMETRY.
CD 78. ..TRIANGULAR, RECTANGULAR BOUNDARY, QUARTER
CD CORE SYMMETRY.
CD 80...TRIANGULAR, RECTANGULAR BOUNDARY, FULL CORE.
CD 90...TRIANGULAR-Z, RHOMBIC BOUNDARY IN PLANE, CORE
CD CENTER LINE AT 60 DEGREE ANGLE.
CD 92...TRIANGULAR-Z, RECTANGULAR BOUNDARY IN PLANE,
CD HALF CORE SYMMETRY IN PLANE.
CD 94...TRIANGULAR-Z, RHOMBIC BOUNDARY IN PLANE, CORE
CD CENTER LINE AT 120 DEGREE ANGLE.
157
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
C
CN
CN
C
C
96...TRIANGULAR-Z , 60 DEGREE TRIANGULAR BOUNDARY
IN PLANE.
98...TRIANGULAR-Z , RECTANGULAR BOUNDARY IN PLANE,
QUARTER CORE SYMMETRY IN PLANE.
100...TRIANGULAR-Z , RECTANGULAR BOUNDARY IN PLANE,
FULL CORE IN PLANE.
110...HEXAGONAL, FULL CORE.
114...HEXAGONAL, SIXTH CORE SYMMETRY.
116...HEXAGONAL, THIRD CORE SYMMETRY.
120...HEXAGONAL-Z , FULL CORE IN PLANE.
124...HEXAGONAL-Z , SIXTH CORE SYMMETRY IN PLANE.
126...HEXAGONAL-Z , THIRD CORE SYMMETRY IN PLANE.
THE HEXAGONAL AND HEXAGONAL-Z GEOMETRY OPTIONS MAY
NOT BE AVAILABLE IN ALL VERSIONS OF DIF3D.
C
CR
C
CL
C
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
CD
EXTERNAL BOUNDARY CONDITIONS (TYPE 04)
FORMAT (I2,10X,616)
COLUMNS CONTENTS...IMPLICATIONS, IF ANY
= = = = = = = SS= SS33S= 3SX3= 3SS33= = 33S3S= S3X3Sra= = 33= SSS= 3= X3SSSSS:
1-2 04
13-18 BOUNDARY CONDITION AT LOWER "X" BOUNDARY OF REACTOR.
19-24 BOUNDARY CONDITION AT UPPER "X" BOUNDARY OF REACTOR.
25-30 BOUNDARY CONDITION AT LOWER "Y" BOUNDARY OF REACTOR.
31-36 BOUNDARY CONDITION AT UPPER "Y" BOUNDARY OF REACTOR.
37-42 BOUNDARY CONDITION AT LOWER Z BOUNDARY OF REACTOR.
43-48 BOUNDARY CONDITION AT UPPER Z BOUNDARY OF REACTOR.
2...PHI-0.
3...PHI PRIME-0.
4...D * PHI PRIME + A * PHI - 0.
6...REPEATING (PERIODIC) WITH OPPOSITE FACE.
7...REPEATING (PERIODIC) WITH NEXT ADJACENT BOUNDARY
(SEE DISCUSSION BELOW).
8...INVERTED REPEATING ALONG THIS FACE
(180 DEGREE ROTATION).
9...INCOMING ANGULAR FLUX Z ERO (TRANSPORT ONLY).
10..REFLECTIVE (TRANSPORT ONLY).
158
CD 11..PERIODIC (TRANSPORT O N L Y ).
CD 12..WHITE (TRANSPORT O N L Y ).
CD
C
CN PHI PRIME IS THE DERIVATIVE OF THE FLUX IN THE
CN DIRECTION OF THE REACTOR OUTWARD NORMAL. D IS THE
CN DIFFUSION COEFFICIENT IN THE MESH INTERVAL
CN IMMEDIATELY INSIDE THE REACTOR BOUNDARY. IF COLS.
CN 43-48 ARE 4 AND NO TYPE 05 CARD IS SUPPLIED TO SPECIFY
CN THE CONSTANT A, THE VALUE 0.46920 WILL BE USED BY
CN DEFAULT.
CN
CN CONDITIONS 2-8 APPLY TO DIFFUSION THEORY PROBLEMS,
CN AND 9-12 APPLY TO TRANSPORT THEORY PROBLEMS.
CN
CN "X" REPRESENTS THE FIRST DIMENSION COORDINATE (X IN
CN X-Y GEOMETRY, R IN R-Z, E T C . ). "Y" REPRESENTS THE
CN SECOND DIMENSION COORDINATE (Y IN X-Y GEOMETRY, Z IN
CN R - Z, E T C . ). WHEN THE MODEL IS THREE-DIMENSIONAL, THE
CN THIRD DIMENSION IS ALWAYS Z.
CN
CN REPEATING CONDITIONS (6,7,8) ARE ONLY APPLICABLE TO
CN THE FIRST TWO DIMENSIONS.
CN
CN NOTE FOR REPEATING CONDITION 7. LET XL DENOTE THE
CN LOWER "X" BOUNDARY, XU DENOTE THE UPPER "X" BOUNDARY,
CN YL DENOTE THE LOWER "Y" BOUNDARY AND YU DENOTE THE
CN UPPER Y BOUNDARY. FOR REPEATING BOUNDARY CONDITIONS
CN (CONDITION 7 ), THE SEQUENCE OF BOUNDARIES IMPLIED BY
CN THE TERM "NEXT ADJACENT BOUNDARY" IS XL, Y L , XU, YU.
CN OF THE TWO BOUNDARIES INVOLVED, THE ONE APPEARING
CN FIRST IN THE SEQUENCE IS ASSIGNED THE BOUNDARY
CN CONDITION (7 ), THE SECOND IS IGNORED. FOR EXAMPLE,
CN IF XL AND YL ARE THE PERIODIC BOUNDARIES, COLS. 13-18
CN MUST CONTAIN A 7, COLS. 25-30 WILL BE IGNORED.
C
CR VARIABLE-MESH STRUCTURE (TYPE 09)
C
CL FORMAT (I2,9X,A1,3(I6,E12.5))
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD 1-2 09
CD
CD 12 COORDINATE DIRECTION.
CD X..."X" COORDINATE DIRECTION.
CD Y..."Y
M
COORDINATE DIRECTION.
CD Z...Z-COORDINATE DIRECTION.
159
CD
CD 13-18 NUMBER OF INTERVALS.
CD
CD 19-30 UPPER COORDINATE.
CD
CD 31-36 NUMBER OF INTERVALS.
C
CD 37-48 UPPER COORDINATE.
CD
CD 49-54 NUMBER OF INTERVALS.
CD
CD 55-66 UPPER COORDINATE.
C
CN NOTE THAT A Z IN COL. 12 IS PERTINENT ONLY IF THE
CN GEOMETRY IS THREE-DIMENSIONAL.
CN
CN "X" REPRESENTS THE FIRST DIMENSION COORDINATE (X IN
CN X-Y GEOMETRY, R IN R-Z , ET C .) . "Y" REPRESENTS THE
CN SECOND DIMENSION COORDINATE (Y IN X-Y GEOMETRY, Z IN
CN R-Z , ET C .) . WHEN THE MODEL IS THREE-DIMENSIONAL, THE
CN THIRD DIMENSION IS ALWAYS Z .
C
CN IN GEOMETRIES INVOLVING AN ANGULAR DIMENSION (THETA)
CN THE ANGULAR VARIABLE MUST BE GIVEN IN RADIANS.
CN
CN EACH NUMBER PAIR IS OF THE FORM (N(I), X( I) ) . THERE
CN ARE N( I) INTERVALS BETWEEN X(I-l) AND X( I) , WHERE X(O)
CN IS THE LOWER REACTOR BOUNDARY IN THIS DIRECTION.
CN NUMBER PAIRS MUST BE GIVEN IN ORDER OF INCREASING
CN MESH COORDINATES. ALL REGION BOUNDARIES MUST COINCIDE
CN WITH THE MESH LINES THAT BOUND MESH INTERVALS.
C
c
CR LOCATIONS OF REGIONS FOR TRIANGULAR, TRIANGULAR-Z ,
CR HEXAGONAL, AND HEXAGONAL-Z GEOMETRIES (TYPE 30)
C
CL FORMAT (12,4X,A6,316,2E12.5)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD 1-2 30
CD
CD 7-12 REGION LABEL (REPEATED ON ADDITIONAL TYPE 30 C AR D S) .
CD
CD 13-18 HEXAGONAL RING NUMBER WHERE REGION IS LOCATED.
CD
CD 19-24 STARTING HEXAGON POSITION FOR THIS REGION.
CD
CD 25-30 FINAL HEXAGON POSITION FOR THIS REGION.
160
CD
CD 31-42 LOWER Z BOUNDARY OF REGION.
CD
CD 43-54 UPPER 2 BOUNDARY OF REGION.
CD
C
CN REGION LABELS MUST BE NON-BLANK.
CN
CN IF THE STARTING POSITKtt (COLS. 19-24) IS BLANK OR
CN Z ERO, THE REGION LABEL U ASSIGNED TO THE WHOLE RING.
CN
CN IF THE FINAL POSITION (COLS. 25-30) IS BLANK OR Z ERO,
CN THE REGION LABEL IS ASSIGNED TO THE POSITION IN 19-24
CN OF THE RING IN 13-18.
CN
CN DATA ON THIS CARD MAY BE OVERLAYED. THAT IS, REGION
CN ASSIGNMENTS DEFINED ON LATER TYPE 30 CARDS SUPERCEDE
CN DATA FOR RINGS AND POSITIONS PREVIOUSLY SPECIFIED.
CN
CN THE REGION LOWER AND UPPER Z BOUNDARIES MUST COINCIDE
CN WITH MESH LINES, WHICH BOUND MESH INTERVALS.
CN
CN THE FIGURE BELOW ILLUSTRATES THE ORDER OF NAMING
CN RINGS AND HEXAGONS IN THE RINGS. THE FIRST NUMBER OF
CN EACH NUMBERED PAIR IS THE RING NUMBER, AND THE SECOND
CN NUMBER IS THE HEXAGON NUMBER IN THAT RING.
CN THE REGION OF SOLUTION DEPENDS ON THE VALUE IN COLS.
CN 13-18 ON CARD TYPE 03 AS FOLLOWS.
CN
CN COLS. 13-18 ON CARD TYPE 03 REGION OF SOLUTION
CN 80 ENTIRE FIGURE AS SHOWN BELOW
CN 72 IN THE 180 DEGREE SECTOR A-B
CN 78 IN THE 90 DEGREE SECTOR A-C
CN 70 IN THE 60 DEGREE SECTOR A-D
CN 74 IN THE 120 DEGREE SECTOR A-E
CN 110 ENTIRE FIGURE AS SHOWN BELOW
CN 114 IN THE 60 DEGREE SECTOR F-C
CN 116 IN THE 120 DEGREE SECTOR F-G
161
CN
CN G E C
CN
CN * * *
CN
CN B * * *
CN * D
CN * * * * *
C
N * ...... ...... ...... *
CN * -3,5 - -3,4 - -3,3 = *
CN * - ...... ...... *
CN ..- ...... ...... ......
CN -3,6 - -2,3 - -2,2 - -3,2 -
CN ...... ...... .....x . . . a s .
CN ...... ...... ...... ......
CN -3,7 - -2,4 - -1,1 - -2,1 - -3,1 - * * * F
CN ... ...... ...... ...... ......
CN ...... ...... ......
CN -3,8 - -2,5 - -2,6 - - 3,12-
CN ..... ..a ... ...... ......
CN ... ...... ...... *
CN -3,9 - -3,10- -3,11- *
CN ...... ...... ...... *
CN *
CN A
CN
CN
CN ALTHOUGH THE REGIONS OF SOLUTION DIFFER FOR THE
CN TRIANGULAR AND HEXAGONAL GEOMETRY MODELS, TYPE 30
CN CARDS COMPOSED FOR TRIANGULAR GEOMETRY MODELS CAN ALSO
CN BE USED FOR HEXAGONAL GEOMETRY MODELS.
CN
C
~-
CR BACKGROUND REGION NAME FOR TRIANGULAR, TRIANGULAR-Z ,
CR HEXAGONAL, AND HEXAGONAL-Z GEOMETRIES (TYPE 31)
C
CL FORMAT (I2,4X,A6)
C
CD COLUMNS CONTENTS...IMPLICATIONS, IF ANY
CD ....... ...................................a................!
CD 1-2 31
CD
CD 7-12 BACKGROUND REGION NAME.
C
CN ANY PORTION OF THE REACTOR NOT SPECIFIED ON THE
CN TYPE 30 CARDS WILL BE IN THE BACKGROUND REGION,
CN
CN IF THE BACKGROUND REGION NAME (COLS. 7-12) IS BLANK,
162
CN OR IF THERE IS NO TYPE 31 CARD, THE BACKGROUND REGION
CN WILL BE ASSIGNED A REGION NUMBER 0 (Z ERO). NOTE THAT
CN SOME CCCC CODES EXCLUDE SUCH A REGION FROM THE REGION
CN OF SOLUTION, WHILE OTHER CCCC CODES MAY NOT ALLOW
CN Z ERO REGION NUMBERS.
C
163
Distribution for ANL-83-1
Internal;
p> B. Abramson
C< H. Adams
P. I. Amundson
C* L. Beck
E. Sg Beckjord
J. C. Beitel
S. K. Bhattacharyya
R. N. Blomquist
M> M> Bretscher
H. Bigelaw
S> B> Brumbach
R> G. Bucher
J. E. Cahalan
S> G> Carpenter
B. R. Chandler
Y. I. Chang
P. J. Collins
R> J> Cornelia
D. C Cutforth
T. A. Daly
J. R. Deen
K. L. Derstine
D. R> Ferguson
K. E. Freese
E. K. Fujita
P> J> Garner
J. M. Gasidlo
E. M. Gelbard
G> M> Greenman
G> L> Grasseschi
H> Henryson
H> H> Hummel
R> N> Hwang
R. E. Kaiser
Kalimullah
H. Khalil
R> D> Lawrence (5)
W. K. Lehto
R. M. Lell
L. G. LeSage
J> R> Liaw
M> J> Lineberry
D. J. Malloy
J. E. Matos
H. F. McFarlane
R. D. McKnight
D. Meneghetti
A> Olson
Y> Orechwa
E> M> Pennington
P. J> Persian!
P> A. Pizzica
R. B. Pond
j> R> R OSS
R. R. Rudolph
G. K. Rusch
R. W. Schaefer
D. M. Smith
K. S. Smith
J> L> Snelgrove
C. G. Stenberg
W> J> Sturm
S> F> SU
C. E. Till
B. J. Toppel
A. Travelli
R. B. Turski
A. J. Ulrich
R. Vilim
D. C. Wade
D. P. Weber
T. Y. Wei
W. L. Woodruff
S. T. Yang
B. S. Yarlagadda
ANL Patent Dept.
ANL Contract File
ANL Libraries (2)
T1S Files (6)
AP Division
Files (10)
External:
DOE-TIC, for distribution per UC-79d (122)
Manager, Chicago Operations Office, DOE
Director, Technology Management, DOE-CH
Director, DOE-RRT (2)
Applied Physics Division Review Committee:
P. W. Dickson, Jr., Clinch River Breeder Reactor Project, Oak Ridge
K> D. Lathrop, Los Alamos National Laboratory
D. A> Keneley, Ontario Hydro
J. E> Meyer, Massachusetts Inst. Technology
R. Sher, Stanford U.
D . B. Wehmeyer, Detroit Edison
A. E. Wilson, Idaho State U.
H. Alter, Office of Breeder Technology, DOS
Advanced Reactor Library, Westinghouse Electric Co., Madison, Pa
M. Becker, Rensselaer Polytechnic Inst.
R. A. Bennett, Westinghouse Hanford Co.
164
C. Cowan, General Electric Co., Sunnyvale
Fast Breeder Dept. Library, General Electric Co., Sunnyvale
H. L. Dodds, Technology for Energy Corp., Knoxville
R. Doncals, Westinghouse Electric Corp., Madison, Pa
J. J. D o mi n g , University of Illinois, Urbana
M. J. Driscoll, MIT
C. Durston, Combustion Engineering, Windsor
R. Ehrlich, General Electric Co., San Jose
H. Farrar IV, Atomics International, Canoga Park
G. F. Flanagan, Oak Ridge National Laboratory
N. M. Greene, Oak Ridge National Laboratory
D. R. Harris, Rensselaer Polytechnic Inst.
P. B. Hemmig, Reactor Research and Development, DOE
A. F. Henry, MIT
J. Kallfelz, Georgia Institute of Technology
R. Karam, Georgia Institute of Technology
W. Y. Kato, Brookhaven National Laboratory
R. J. LaBauve, Los Alamos National Laboratory
J. Lewellen, Reactor Research and Development, DOE
E. Lewis, Northwestern University
M. D. Libby, NUSCO, Hartford, CT
D. R. McCoy, Los Alamos National Laboratory
F. C. Maienschein, Oak Ridge National Laboratory
D. R. Mathews, GA Technologies, San Diego
R. MacFarlane, Los Alamos National Laboratory
H. A. Morowitz, Tarzana, CA
J. Naser, Electric Power Research Inst., Palo Alto
National Energy Software Center, ANL (10)
R. J. Neuhold, Reactor Research and Development, DOE
R. D. O'Dell, Los Alamos National Laboratory
D. Okrent, University of California, Los Angeles
K. Ott, Purdue University
0. Ozer, Electric Power Research Inst., Palo Alto
A. M. Perry, Oak Ridge National Laboratory
J. Prabulos, Combustion Engineering, Windsor
Radiation Shielding Information Center, ORNL (10)
A. B. Reynolds, University of Virginia
P. Rose, Brookhaven National Laboratory
D. H. Roy, Babcock and Wilcox Co., Lynchburg, Va
R. Schenter, Westinghouse Hanford Co.
P. Soran, Los Alamos National Laboratory
E. R. Specht, Atomics International, Canoga Park
S. Stewart, General Electric, Co., Sunnyvale
L. E. Strawbridge, Westinghouse Electric Corp., Madison, Pa
R. J. Tuttle, Atomics International, Canoga Park
D. R. Vondy, Oak Ridge National Laboratory
C. Weisben, Oak Ridge National Laboratory